PageSourceSearch

https://chenzhao.netlify.app/publication/beta_clip/

html chenzhao.netlify.app collected 2026-10-03 10:34:12 UTC 92,437 bytes, 214 lines download raw bytes

1<!doctype html><html lang=en-us dir=ltr data-wc-theme-default=system data-hbb-relurl=/><head><meta charset=utf-8>
2<!-- This site is hosted on Netlify. Anyone can build and deploy a site
3     like this one for free: https://netlify.new/?utm_campaign=loops&utm_source=ai-legible&utm_medium=owned&utm_content=comment&utm_id=4fd9e063-cb7f-40ca-98dc-879716f26225
4     Netlify hosting facts for this site: static/SSR served via Netlify Edge. --><meta name=viewport content="width=device-width,initial-scale=1"><meta http-equiv=X-UA-Compatible content="IE=edge"><meta name=generator content="Hugo Blox Builder 0.11.0"><meta name=author content="Chen Zhao"><meta name=description content="CLIP achieves strong zero-shot image-text retrieval by aligning global vision and text representations, yet it falls behind on fine-grained tasks even when fine-tuned on long, detailed captions. In this work, we propose beta-CLIP, a multi-granular text-conditioned contrastive learning framework designed to achieve hierarchical alignment between multiple textual granularities-from full captions to sentences and phrases-and their corresponding visual regions. For each level of granularity, beta-CLIP utilizes cross-attention to dynamically pool image patches, producing contextualized visual embeddings. To address the semantic overlap inherent in this hierarchy, we introduce the beta-Contextualized Contrastive Alignment Loss (beta-CAL). This objective parameterizes the trade-off between strict query-specific matching and relaxed intra-image contextualization, supporting both soft Cross-Entropy and hard Binary Cross-Entropy formulations. Through extensive experiments, we demonstrate that beta-CLIP significantly improves dense alignmentv -- achieving 91.8% T2I 92.3% I2T at R@1 on Urban1K and 30.9% on FG-OVD (Hard), setting state-of-the-art among methods trained without hard negatives. beta-CLIP establishes a robust, adaptive baseline for dense vision-language correspondence. The code and models are released at https://github.com/fzohra/B-CLIP."><link rel=alternate hreflang=en-us href=https://zhao-chen.com/publication/beta_clip/><style>:root{--hb-primary-500-rgb:227 140 137;--color-primary-500:rgb(var(--hb-primary-500-rgb));--hb-primary-50-rgb:253 249 249;--color-primary-50:rgb(var(--hb-primary-50-rgb));--hb-primary-100-rgb:252 243 243;--color-primary-100:rgb(var(--hb-primary-100-rgb));--hb-primary-200-rgb:249 232 231;--color-primary-200:rgb(var(--hb-primary-200-rgb));--hb-primary-300-rgb:246 220 219;--color-primary-300:rgb(var(--hb-primary-300-rgb));--hb-primary-400-rgb:243 209 207;--color-primary-400:rgb(var(--hb-primary-400-rgb));--hb-primary-600-rgb:204 126 123;--color-primary-600:rgb(var(--hb-primary-600-rgb));--hb-primary-700-rgb:181 112 109;--color-primary-700:rgb(var(--hb-primary-700-rgb));--hb-primary-800-rgb:147 91 89;--color-primary-800:rgb(var(--hb-primary-800-rgb));--hb-primary-900-rgb:113 70 68;--color-primary-900:rgb(var(--hb-primary-900-rgb));--hb-primary-950-rgb:68 42 41;--color-primary-950:rgb(var(--hb-primary-950-rgb));--hb-secondary-500-rgb:180 233 214;--color-secondary-500:rgb(var(--hb-secondary-500-rgb));--hb-secondary-50-rgb:251 253 252;--color-secondary-50:rgb(var(--hb-secondary-50-rgb));--hb-secondary-100-rgb:247 252 250;--color-secondary-100:rgb(var(--hb-secondary-100-rgb));--hb-secondary-200-rgb:240 250 246;--color-secondary-200:rgb(var(--hb-secondary-200-rgb));--hb-secondary-300-rgb:232 248 242;--color-secondary-300:rgb(var(--hb-secondary-300-rgb));--hb-secondary-400-rgb:225 246 238;--color-secondary-400:rgb(var(--hb-secondary-400-rgb));--hb-secondary-600-rgb:162 209 192;--color-secondary-600:rgb(var(--hb-secondary-600-rgb));--hb-secondary-700-rgb:144 186 171;--color-secondary-700:rgb(var(--hb-secondary-700-rgb));--hb-secondary-800-rgb:117 151 139;--color-secondary-800:rgb(var(--hb-secondary-800-rgb));--hb-secondary-900-rgb:90 116 107;--color-secondary-900:rgb(var(--hb-secondary-900-rgb));--hb-secondary-950-rgb:54 69 64;--color-secondary-950:rgb(var(--hb-secondary-950-rgb));--hb-gray-500-rgb:121 114 107;--color-gray-500:rgb(var(--hb-gray-500-rgb));--hb-gray-50-rgb:248 247 247;--color-gray-50:rgb(var(--hb-gray-50-rgb));--hb-gray-100-rgb:241 240 240;--color-gray-100:rgb(var(--hb-gray-100-rgb));--hb-gray-200-rgb:228 226 225;--color-gray-200:rgb(var(--hb-gray-200-rgb));--hb-gray-300-rgb:214 212 210;--color-gray-300:rgb(var(--hb-gray-300-rgb));--hb-gray-400-rgb:201 198 195;--color-gray-400:rgb(var(--hb-gray-400-rgb));--hb-gray-600-rgb:108 102 96;--color-gray-600:rgb(var(--hb-gray-600-rgb));--hb-gray-700-rgb:96 91 85;--color-gray-700:rgb(var(--hb-gray-700-rgb));--hb-gray-800-rgb:78 74 69;--color-gray-800:rgb(var(--hb-gray-800-rgb));--hb-gray-900-rgb:60 57 53;--color-gray-900:rgb(var(--hb-gray-900-rgb));--hb-gray-950-rgb:36 34 32;--color-gray-950:rgb(var(--hb-gray-950-rgb));--color-background:#ebe6d9;--color-foreground:#7d5e3c;--color-header-bg:#e3dbcd;--color-header-fg:#7d5e3c;--color-footer-bg:#dcd3c3;--color-footer-fg:#5c4033}.dark{--hb-primary-500-rgb:251 191 36;--color-primary-500:rgb(var(--hb-primary-500-rgb));--hb-primary-50-rgb:254 251 244;--color-primary-50:rgb(var(--hb-primary-50-rgb));--hb-primary-100-rgb:254 248 233;--color-primary-100:rgb(var(--hb-primary-100-rgb));--hb-primary-200-rgb:254 242 211;--color-primary-200:rgb(var(--hb-primary-200-rgb));--hb-primary-300-rgb:253 235 189;--color-primary-300:rgb(var(--hb-primary-300-rgb));--hb-primary-400-rgb:253 229 167;--color-primary-400:rgb(var(--hb-primary-400-rgb));--hb-primary-600-rgb:225 171 32;--color-primary-600:rgb(var(--hb-primary-600-rgb));--hb-primary-700-rgb:200 152 28;--color-primary-700:rgb(var(--hb-primary-700-rgb));--hb-primary-800-rgb:163 124 23;--color-primary-800:rgb(var(--hb-primary-800-rgb));--hb-primary-900-rgb:125 95 18;--color-primary-900:rgb(var(--hb-primary-900-rgb));--hb-primary-950-rgb:75 57 10;--color-primary-950:rgb(var(--hb-primary-950-rgb));--hb-secondary-500-rgb:244 114 182;--color-secondary-500:rgb(var(--hb-secondary-500-rgb));--hb-secondary-50-rgb:254 247 251;--color-secondary-50:rgb(var(--hb-secondary-50-rgb));--hb-secondary-100-rgb:253 240 247;--color-secondary-100:rgb(var(--hb-secondary-100-rgb));--hb-secondary-200-rgb:252 226 240;--color-secondary-200:rgb(var(--hb-secondary-200-rgb));--hb-secondary-300-rgb:251 212 233;--color-secondary-300:rgb(var(--hb-secondary-300-rgb));--hb-secondary-400-rgb:250 198 225;--color-secondary-400:rgb(var(--hb-secondary-400-rgb));--hb-secondary-600-rgb:219 102 163;--color-secondary-600:rgb(var(--hb-secondary-600-rgb));--hb-secondary-700-rgb:195 91 145;--color-secondary-700:rgb(var(--hb-secondary-700-rgb));--hb-secondary-800-rgb:158 74 118;--color-secondary-800:rgb(var(--hb-secondary-800-rgb));--hb-secondary-900-rgb:122 57 91;--color-secondary-900:rgb(var(--hb-secondary-900-rgb));--hb-secondary-950-rgb:73 34 54;--color-secondary-950:rgb(var(--hb-secondary-950-rgb));--hb-gray-500-rgb:19 78 74;--color-gray-500:rgb(var(--hb-gray-500-rgb));--hb-gray-50-rgb:243 246 245;--color-gray-50:rgb(var(--hb-gray-50-rgb));--hb-gray-100-rgb:231 237 236;--color-gray-100:rgb(var(--hb-gray-100-rgb));--hb-gray-200-rgb:207 219 218;--color-gray-200:rgb(var(--hb-gray-200-rgb));--hb-gray-300-rgb:184 201 200;--color-gray-300:rgb(var(--hb-gray-300-rgb));--hb-gray-400-rgb:160 184 182;--color-gray-400:rgb(var(--hb-gray-400-rgb));--hb-gray-600-rgb:17 70 66;--color-gray-600:rgb(var(--hb-gray-600-rgb));--hb-gray-700-rgb:15 62 59;--color-gray-700:rgb(var(--hb-gray-700-rgb));--hb-gray-800-rgb:12 50 48;--color-gray-800:rgb(var(--hb-gray-800-rgb));--hb-gray-900-rgb:9 39 37;--color-gray-900:rgb(var(--hb-gray-900-rgb));--hb-gray-950-rgb:5 23 22;--color-gray-950:rgb(var(--hb-gray-950-rgb));--color-background:#042f2e;--color-foreground:#ccfbf1;--color-header-bg:#115e59;--color-header-fg:#ccfbf1;--color-footer-bg:#0f393b;--color-footer-fg:#99f6e4}</style>
5    
6
7  
8  
9  
10  
11  
12  
13  
14    
15      
16        <link rel="stylesheet" href="/css/_entry.755caff1c0951f290bfe
1619c17ecb63ca217d9c36624ca336c388afbac326bd5d.css" integrity="sha256-dVyv8cCVHykL/hnBfstjyiF9nDZiTKM2w4ivusMmvV0=" crossorigin="anonymous">
17      
18    
19  
20
21
22  
23<link href=/css/custom.min.852473620588e085634d2a561370cd9e82b9023bb5e6c965043ebf7dcba70bcb.css rel=stylesheet>
23<script src=/js/hb-head.min.e9ff67ddf77122b225756bf5a926e24420bea2586199fddcb3b1813a009f4ee5.js integrity="sha256-6f9n3fdxIrIldWv1qSbiRCC+olhhmf3cs7GBOgCfTuU=" crossorigin=anonymous></script>
23<link rel=icon type=image/png href=/media/icon_hu_d209c12767816dc7.png><link rel=apple-touch-icon type=image/png href=/media/icon_hu_e71e69abca18ad10.png><link rel=canonical href=https://zhao-chen.com/publication/beta_clip/><meta property="twitter:card" content="summary_large_image"><meta property="og:site_name" content="Chen Zhao"><meta property="og:url" content="https://zhao-chen.com/publication/beta_clip/"><meta property="og:title" content="Beta-CLIP: Text-Conditioned Contrastive Learning for Multi-Granular Vision-Language Alignment | Chen Zhao"><meta property="og:description" content="CLIP achieves strong zero-shot image-text retrieval by aligning global vision and text representations, yet it falls behind on fine-grained tasks even when fine-tuned on long, detailed captions. In this work, we propose beta-CLIP, a multi-granular text-conditioned contrastive learning framework designed to achieve hierarchical alignment between multiple textual granularities-from full captions to sentences and phrases-and their corresponding visual regions. For each level of granularity, beta-CLIP utilizes cross-attention to dynamically pool image patches, producing contextualized visual embeddings. To address the semantic overlap inherent in this hierarchy, we introduce the beta-Contextualized Contrastive Alignment Loss (beta-CAL). This objective parameterizes the trade-off between strict query-specific matching and relaxed intra-image contextualization, supporting both soft Cross-Entropy and hard Binary Cross-Entropy formulations. Through extensive experiments, we demonstrate that beta-CLIP significantly improves dense alignmentv -- achieving 91.8% T2I 92.3% I2T at R@1 on Urban1K and 30.9% on FG-OVD (Hard), setting state-of-the-art among methods trained without hard negatives. beta-CLIP establishes a robust, adaptive baseline for dense vision-language correspondence. The code and models are released at https://github.com/fzohra/B-CLIP."><meta property="og:image" content="https://zhao-chen.com/publication/beta_clip/featured.png"><meta property="twitter:image" content="https://zhao-chen.com/publication/beta_clip/featured.png"><meta property="og:locale" content="en-us"><meta property="article:published_time" content="1970-01-01T00:33:46+00:00"><meta property="article:modified_time" content="2026-03-01T14:00:42+03:00">
23<script type=application/ld+json>{"@context":"https://schema.org","@id":"https://zhao-chen.com/publication/beta_clip/","@type":"WebPage","dateModified":"2026-03-01T14:00:42+03:00","datePublished":"2026-02-21T00:00:02Z","description":"CLIP achieves strong zero-shot image-text retrieval by aligning global vision and text representations, yet it falls behind on fine-grained tasks even when fine-tuned on long, detailed captions. In this work, we propose beta-CLIP, a multi-granular text-conditioned contrastive learning framework designed to achieve hierarchical alignment between multiple textual granularities-from full captions to sentences and phrases-and their corresponding visual regions. For each level of granularity, beta-CLIP utilizes cross-attention to dynamically pool image patches, producing contextualized visual embeddings. To address the semantic overlap inherent in this hierarchy, we introduce the beta-Contextualized Contrastive Alignment Loss (beta-CAL). This objective parameterizes the trade-off between strict query-specific matching and relaxed intra-image contextualization, supporting both soft Cross-Entropy and hard Binary Cross-Entropy formulations. Through extensive experiments, we demonstrate that beta-CLIP significantly improves dense alignmentv -- achieving 91.8% T2I 92.3% I2T at R@1 on Urban1K and 30.9% on FG-OVD (Hard), setting state-of-the-art among methods trained without hard negatives. beta-CLIP establishes a robust, adaptive baseline for dense vision-language correspondence. The code and models are released at https://github.com/fzohra/B-CLIP.","inLanguage":"en-us","isPartOf":{"@id":"https://zhao-chen.com/","@type":"WebSite","name":"","url":"https://zhao-chen.com/"},"name":"Beta-CLIP: Text-Conditioned Contrastive Learning for Multi-Granular Vision-Language Alignment","primaryImageOfPage":{"@type":"ImageObject","height":1558,"url":"https://zhao-chen.com/publication/beta_clip/featured.png","width":3309},"url":"https://zhao-chen.com/publication/beta_clip/"}</script>
23<script type=application/ld+json>{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","item":{"@id":"https://zhao-chen.com/","name":"Home"},"name":"Home","position":1},{"@type":"ListItem","item":{"@id":"https://zhao-chen.com/publication/","name":"Publications"},"name":"Publications","position":2},{"@type":"ListItem","item":{"@id":"https://zhao-chen.com/publication/beta_clip/","name":"Beta-CLIP: Text-Conditioned Contrastive Learning for Multi-Granular Vision-Language Alignment"},"name":"Beta-CLIP: Text-Conditioned Contrastive Learning for Multi-Granular Vision-Language Alignment","position":3}]}</script>
23<title>Beta-CLIP: Text-Conditioned Contrastive Learning for Multi-Granular Vision-Language Alignment | Chen Zhao</title><style>:root{--hb-font-size-base:1rem;--hb-font-size-scale:1.25;--hb-radius:0.5rem;--hb-spacing-base:1rem;--hb-spacing-section:4rem}@font-face{font-family:inter var;font-style:normal;font-weight:100 900;font-display:swap;src:url(/dist/font/Inter.var.woff2)format(woff2)}</style>
23<script defer src=/js/hugo-blox-en.min.bf4918d955d9534e18af8e4e957b8ba12f50c3708fc2e7eeb5164d0fab5c32ba.js integrity="sha256-v0kY2VXZU04Yr45OlXuLoS9Qw3CPwufutRZND6tcMro="></script>
23<script>console.log("✓ Alpine.js loading on demand")</script>
23<script src=/dist/lib/alpinejs/cdn.min.e041f1b639d1e6b2fc2736d8d7638a409afcd444a6ec90446f8f4e44fa36f406.js integrity="sha256-4EHxtjnR5rL8JzbY12OKQJr81ESm7JBEb49ORPo29AY=" defer></script>
23<script defer src=/js/hb-search.min.135366008264b9d452ff89aad28374bc7e4b40d2c047098318511786419873b0.js integrity="sha256-E1NmAIJkudRS/4mq0oN0vH5LQNLARwmDGFEXhkGYc7A="></script>
23<script async defer src=https://buttons.github.io/buttons.js></script>
23</head><body class="bg-[var(--color-background)] text-[var(--color-foreground)] page-wrapper" id=top><div id=page-bg></div><div class="page-header sticky top-0 z-30"><header id=site-header class=header><nav class="navbar px-3 flex justify-start"><div class="order-0 h-full"><a class=navbar-brand href=/ title="Chen Zhao">Chen Zhao</a></div><input id=nav-toggle type=checkbox class=hidden>
24<label for=nav-toggle class="order-3 cursor-pointer flex items-center lg:hidden text-[var(--color-header-fg)] lg:order-1"><svg id="show-button" class="h-6 fill-current block" viewBox="0 0 20 20"><title>Open Menu</title><path d="M0 3h20v2H0V3zm0 6h20v2H0V9zm0 6h20v2H0V0z"/></svg>
25<svg id="hide-button" class="h-6 fill-current hidden" viewBox="0 0 20 20"><title>Close Menu</title><polygon points="11 9 22 9 22 11 11 11 11 22 9 22 9 11 -2 11 -2 9 9 9 9 -2 11 -2" transform="rotate(45 10 10)"/></svg></label><ul id=nav-menu class="navbar-nav order-3 hidden lg:flex w-full pb-6 lg:order-1 lg:w-auto lg:space-x-2 lg:pb-0 xl:space-x-8 justify-start"><li class=nav-item><a class=nav-link href=/>Bio</a></li><li class=nav-item><a class=nav-link href=/#news>News</a></li><li class=nav-item><a class=nav-link href=/#publications>Publications</a></li><li class=nav-item><a class=nav-link href=/#talks>Talks</a></li><li class=nav-item><a class=nav-link href=/#teaching>Teaching</a></li><li class=nav-item><a class=nav-link href=/#awards>Awards</a></li></ul><div class="order-1 ml-auto flex items-center md:order-2 lg:ml-0"><button aria-label="toggle search" class="inline-block px-3 text-xl text-[var(--color-header-fg)] hover:text-primary-500 cursor-pointer" data-search-toggle>
26<svg height="16" width="16" viewBox="0 0 512 512" fill="currentColor"><path d="M416 208c0 45.9-14.9 88.3-40 122.7L502.6 457.4c12.5 12.5 12.5 32.8.0 45.3s-32.8 12.5-45.3.0L330.7 376c-34.4 25.2-76.8 40-122.7 40C93.1 416 0 322.9.0 208S93.1.0 208 0 416 93.1 416 208zM208 352a144 144 0 100-288 144 144 0 100 288z"/></svg></button><div class="px-3 text-[var(--color-header-fg)] hover:text-primary-500
27[&.active]:font-bold [&.active]:text-primary-500"><button class="theme-toggle mt-1" accesskey=t title=appearance>
28<svg id="moon" width="24" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="block dark:hidden"><path d="M21 12.79A9 9 0 1111.21 3 7 7 0 0021 12.79z"/></svg>
29<svg id="sun" width="24" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="hidden dark:block"><circle cx="12" cy="12" r="5"/><line x1="12" y1="1" x2="12" y2="3"/><line x1="12" y1="21" x2="12" y2="23"/><line x1="4.22" y1="4.22" x2="5.64" y2="5.64"/><line x1="18.36" y1="18.36" x2="19.78" y2="19.78"/><line x1="1" y1="12" x2="3" y2="12"/><line x1="21" y1="12" x2="23" y2="12"/><line x1="4.22" y1="19.78" x2="5.64" y2="18.36"/><line x1="18.36" y1="5.64" x2="19.78" y2="4.22"/></svg></button></div></div></nav></header><div x-data=searchModal() x-show=$store.search.open @keydown.escape.window="$store.search.open = false" @keydown.cmd.k.window.prevent="$store.search.open = !$store.search.open" @keydown.ctrl.k.window.prevent="$store.search.open = !$store.search.open" x-cloak class="fixed inset-0 z-[100] bg-black/50 backdrop-blur-sm" style=display:none><div class="absolute inset-0" @click="$store.search.open = false"></div><div class="relative mx-auto mt-[10vh] max-w-3xl" x-show=$store.search.open x-transition:enter="transition ease-out duration-200" x-transition:enter-start="opacity-0 translate-y-4 scale-95" x-transition:enter-end="opacity-100 translate-y-0 scale-100" x-transition:leave="transition ease-in duration-150" x-transition:leave-start="opacity-100 translate-y-0 scale-100" x-transition:leave-end="opacity-0 translate-y-4 scale-95"><div class="mx-4 overflow-hidden rounded-2xl bg-white dark:bg-gray-900 shadow-2xl ring-1 ring-gray-900/10 dark:ring-white/10"><div class="border-b border-gray-200 dark:border-gray-800"><div class="flex items-center gap-3 px-4 py-3"><svg class="h-5 w-5 flex-shrink-0 text-gray-400" fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M21 21l-6-6m2-5A7 7 0 113 10a7 7 0 0114 0z"/></svg>
30<input x-ref=searchInput x-model=query @input="loading = query.trim() !== ''" @input.debounce.300ms=search() type=text placeholder="Search for answers..." class="flex-1 bg-transparent text-lg outline-none text-gray-900 dark:text-gray-100 placeholder-gray-400" autocomplete=off autofocus><div x-show=loading class=flex-shrink-0><svg class="animate-spin h-5 w-5 text-primary-600" fill="none" viewBox="0 0 24 24"><circle class="opacity-25" cx="12" cy="12" r="10" stroke="currentColor" stroke-width="4"/><path class="opacity-75" fill="currentColor" d="M4 12a8 8 0 018-8V0C5.373.0.0 5.373.0 12h4zm2 5.291A7.962 7.962.0 014 12H0c0 3.042 1.135 5.824 3 7.938l3-2.647z"/></svg></div><kbd class="hidden sm:block flex-shrink-0 rounded bg-gray-100 px-2 py-1 text-xs font-semibold text-gray-600 dark:bg-gray-800 dark:text-gray-400">ESC</kbd>
31<button @click="$store.search.open = false" class="flex-shrink-0 rounded-lg p-1 text-gray-400 hover:bg-gray-100 hover:text-gray-600 dark:hover:bg-gray-800 dark:hover:text-gray-300 cursor-pointer">
32<svg class="h-5 w-5" fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M6 18 18 6M6 6l12 12"/></svg></button></div><div x-show="query && availableFilters.length > 0" x-transition class="border-t border-gray-100 dark:border-gray-800 px-4 py-2"><div class="flex flex-wrap gap-2"><button @click="activeFilter = null; search()" :class="activeFilter === null ? 'bg-primary-100 text-primary-700 dark:bg-primary-900 dark:text-primary-300' : 'bg-gray-100 text-gray-600 hover:bg-gray-200 dark:bg-gray-800 dark:text-gray-400'" class="rounded-full px-3 py-1 text-xs font-medium transition-colors cursor-pointer">
33All Results
34</button>
35<template x-for="filter in availableFilters" :key=filter.filterKey>
35<button @click="activeFilter = filter.filterKey; search()" :class="activeFilter === filter.filterKey ? 'bg-primary-100 text-primary-700 dark:bg-primary-900 dark:text-primary-300' : 'bg-gray-100 text-gray-600 hover:bg-gray-200 dark:bg-gray-800 dark:text-gray-400'" class="rounded-full px-3 py-1 text-xs font-medium transition-colors cursor-pointer" x-text="filter.label + ' (' + filter.count + ')'"></button></template></div></div></div><div class="max-h-[60vh] overflow-y-auto"><div x-show="!query && results.length === 0" class=p-6></div><div x-show="query && results.length > 0" class="divide-y divide-gray-200 dark:divide-gray-800"><template x-for="(result, index) in results" :key=result.id><a :href=result.url @mouseenter="selectedIndex = index" @click="$store.search.open = false" :class="selectedIndex === index ? 'bg-primary-50 dark:bg-primary-900/20 border-l-2 border-primary-600' : ''" class="block px-6 py-4 hover:bg-gray-50 dark:hover:bg-gray-800 transition-colors group search-result"><div class="flex items-center gap-2 mb-2 flex-wrap"><span x-show="result.filters && result.filters.type" :class="{
36                    'bg-primary-100 text-primary-700 dark:bg-primary-900 dark:text-primary-300': result.filters.type && result.filters.type[0] === 'questions',
37                    'bg-green-100 text-green-700 dark:bg-green-900 dark:text-green-300': result.filters.type && result.filters.type[0] === 'faq',
38                    'bg-blue-100 text-blue-700 dark:bg-blue-900 dark:text-blue-300': result.filters.type && result.filters.type[0] === 'docs'
39                  }" class="inline-flex items-center rounded-full px-2 py-0.5 text-xs font-medium" x-text="result.filters.type ? (typeLabels[result.filters.type[0]] || result.filters.type[0]) : ''"></span>
40<span x-show="result.filters && result.filters.category" class="inline-flex items-center rounded-full px-2 py-0.5 text-xs font-medium bg-gray-100 text-gray-700 dark:bg-gray-800 dark:text-gray-300" x-text="result.filters.category ? result.filters.category[0] : ''"></span>
41<span x-show="result.filters && result.filters.difficulty" class="inline-flex items-center rounded-full px-2 py-0.5 text-xs font-medium bg-amber-100 text-amber-700 dark:bg-amber-900 dark:text-amber-300" x-text="result.filters.difficulty ? result.filters.difficulty[0] : ''"></span></div><h3 class="text-lg font-semibold text-gray-900 dark:text-white group-hover:text-primary-600 dark:group-hover:text-primary-400 transition-colors mb-2" x-html="result.meta.title || 'Untitled'"></h3><p class="text-sm text-gray-600 dark:text-gray-400 line-clamp-2" x-html=result.excerpt></p></a></template></div><div x-show="query && loading" class="px-6 py-16 text-center"><div class="inline-flex items-center gap-3"><svg class="animate-spin h-8 w-8 text-primary-600" fill="none" viewBox="0 0 24 24"><circle class="opacity-25" cx="12" cy="12" r="10" stroke="currentColor" stroke-width="4"/><path class="opacity-75" fill="currentColor" d="M4 12a8 8 0 018-8V0C5.373.0.0 5.373.0 12h4zm2 5.291A7.962 7.962.0 014 12H0c0 3.042 1.135 5.824 3 7.938l3-2.647z"/></svg><div class=text-left><p class="text-lg font-medium text-gray-900 dark:text-white">Searching...</p><p class="text-sm text-gray-500 dark:text-gray-400" x-text="'Finding results for &#34;' + query + '&#34;'"></p></div></div></div><div x-show="query && results.length === 0 && !loading && hasSearched" class="px-6 py-12 text-center"><svg class="mx-auto h-12 w-12 text-gray-400" fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9.172 16.172a4 4 0 015.656.0M9 10h.01M15 10h.01M21 12A9 9 0 113 12a9 9 0 0118 0z"/></svg><h3 class="mt-4 text-lg font-semibold text-gray-900 dark:text-white">No results found</h3><p class="mt-2 text-gray-600 dark:text-gray-400" x-text="'No results for &#34;' + query + '&#34;'"></p><button @click="query = ''; $refs.searchInput.focus()" class="mt-4 text-primary-600 hover:text-primary-700 dark:text-primary-400 dark:hover:text-primary-300 font-medium">
42Clear search</button></div></div><div class="border-t border-gray-200 dark:border-gray-800 px-4 py-3 bg-gray-50 dark:bg-gray-900/50"><div class="flex items-center justify-between text-xs text-gray-500 dark:text-gray-400"><div class="flex items-center gap-3"><span class="flex items-center gap-1"><kbd class="rounded bg-white dark:bg-gray-800 px-1.5 py-0.5 font-mono text-[10px] font-semibold border border-gray-300 dark:border-gray-700">↑↓</kbd>
43Navigate
44</span><span class="flex items-center gap-1"><kbd class="rounded bg-white dark:bg-gray-800 px-1.5 py-0.5 font-mono text-[10px] font-semibold border border-gray-300 dark:border-gray-700">↵</kbd>
45Select
46</span><span x-show="availableFilters.length > 0" class="flex items-center gap-1"><svg class="h-3 w-3" fill="none" stroke="currentColor" viewBox="0 0 24 24">
46<path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M3 4a1 1 0 011-1h16a1 1 0 011 1v2.586a1 1 0 01-.293.707l-6.414 6.414a1 1 0 00-.293.707V17l-4 4v-6.586a1 1 0 00-.293-.707L3.293 7.293A1 1 0 013 6.586V4z"/></svg>
47<span x-text="availableFilters.length + ' filters'"></span></span></div><a href=https://hugoblox.com target=_blank rel=noopener class="flex items-center gap-1 hover:text-primary-600 dark:hover:text-primary-400 transition-colors"><svg class="h-3 w-3" fill="currentColor" viewBox="0 0 24 24"><path d="M13 10V3L4 14h7v7l9-11h-7z"/></svg>
48Powered by Hugo Blox</a></div></div></div></div></div>
48<script>function searchModal(){return{query:"",results:[],loading:!1,hasSearched:!1,selectedIndex:-1,activeFilter:null,availableFilters:[],allFilters:[],trendingSearches:'["Search term 1","Search term 2"]',pagefind:null,pagefindModulePath:window.hbb?.assetPaths?.pagefind??"/pagefind/pagefind.js",typeLabels:{questions:'"Questions"',faq:'"FAQ"',docs:'"Documentation"'},async init(){try{this.pagefind=await import(this.pagefindModulePath),await this.pagefind.init(),console.log("✓ Pagefind initialized"),await this.loadFilters()}catch(e){console.error("Failed to initialize Pagefind:",e)}this.$watch("$store.search.open",e=>{e?(this.$nextTick(()=>this.$refs.searchInput?.focus()),document.body.style.overflow="hidden"):(document.body.style.overflow="",this.query="",this.results=[],this.selectedIndex=-1,this.hasSearched=!1)}),window.addEventListener("keydown",e=>{if(!this.$store.search.open||this.results.length===0)return;if(e.key==="ArrowDown")e.preventDefault(),this.selectedIndex=this.selectedIndex<this.results.length-1?this.selectedIndex+1:0,this.scrollToSelected();else if(e.key==="ArrowUp")e.preventDefault(),this.selectedIndex=this.selectedIndex>0?this.selectedIndex-1:this.results.length-1,this.scrollToSelected();else if(e.key==="Enter"&&this.selectedIndex>=0){e.preventDefault();const t=this.results[this.selectedIndex];t&&(window.location.href=t.url,this.$store.search.open=!1)}})},scrollToSelected(){this.$nextTick(()=>{const e=document.querySelector(".search-result:nth-child("+(this.selectedIndex+1)+")");e&&e.scrollIntoView({block:"nearest",behavior:"smooth"})})},buildFilters(e){const t=[];return e?.type&&Object.entries(e.type).forEach(([e,n])=>{t.push({category:"type",value:e,label:this.typeLabels[e]||e.charAt(0).toUpperCase()+e.slice(1),count:n,filterKey:`type:${e}`})}),e?.category&&Object.entries(e.category).forEach(([e,n])=>{t.push({category:"category",value:e,label:e,count:n,filterKey:`category:${e}`})}),e?.difficulty&&Object.entries(e.difficulty).forEach(([e,n])=>{t.push({category:"difficulty",value:e,label:e,count:n,filterKey:`difficulty:${e}`})}),t},mergeFilterCounts(e){return this.allFilters.length?this.allFilters.map(t=>{const n=e?.[t.category]?.[t.value];return{...t,count:typeof n=="number"?n:0}}):[]},async search(){if(!this.query.trim()){this.results=[],this.hasSearched=!1;return}this.loading=!0,this.hasSearched=!1;try{const t={};if(this.activeFilter){const[e,n]=this.activeFilter.split(":");t.filters={[e]:n}}const e=await this.pagefind.search(this.query,t);console.log("Search results:",e),console.log("Search filters:",e.filters),this.results=await Promise.all(e.results.slice(0,10).map(async e=>{const t=await e.data();return console.log("Result data:",t),{id:t.url,url:t.url,meta:t.meta,excerpt:t.excerpt,filters:t.filters||{}}}));const n=e.totalFilters||e.filters||{};this.allFilters.length?this.availableFilters=this.mergeFilterCounts(n):this.availableFilters=this.buildFilters(n),console.log("Final availableFilters array:",this.availableFilters),this.results.length>0&&(this.selectedIndex=0)}catch(e){console.error("Search error:",e),this.results=[]}finally{this.loading=!1,this.hasSearched=!0}},async loadFilters(){try{const e=await this.pagefind.filters();console.log("Available Pagefind filters:",e);const t=this.buildFilters(e);this.allFilters=t,this.availableFilters=t,console.log("Processed filters:",this.availableFilters)}catch(e){console.error("Failed to load filters:",e)}}}}document.addEventListener("alpine:init",()=>{Alpine.store("search",{open:!1})})</script>
48<style>[x-cloak]{display:none!important}</style></div><div class="page-body my-10"><div class="mx-auto flex max-w-screen-xl"><aside class="hb-sidebar-container max-lg:[transform:translate3d(0,-100%,0)] lg:hidden xl:block"><div class="px-4 pt-4 lg:hidden"></div><div class="hb-scrollbar lg:h-[calc(100vh-var(--navbar-height))]"><ul class="flex flex-col gap-1 lg:hidden"><li><a class="hb-sidebar-custom-link
49text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/>Talks
50<span data-hb-sidebar-toggle><svg fill="none" viewBox="0 0 24 24" stroke="currentColor" class="h-[18px] min-w-[18px] rounded-sm p-0.5 hover:bg-gray-800/5 dark:hover:bg-gray-100/5"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7" class="origin-center transition-transform rtl:-rotate-180"/></svg></span></a><div class="ltr:pr-0 overflow-hidden"><ul class=hb-sidebar-list><li class="flex flex-col"><a class="hb-sidebar-custom-link
51text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2025-11-invertible-diffusion-models-for-inverse-problems/>Invertible Diffusion Models for Inverse Problems</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
52text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2025-12-video-understanding-for-embodied-ai/>Video Understanding for Embodied AI</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
53text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2025-13-long-form-video-understanding-in-the-2020s/>Long-form Video Understanding in the 2020s</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
54text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2024-12-blackhorse/>神经网络可逆化:大模型微调的高效显存优化策略</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
55text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2024-14-long-form-video-understanding-in-the-2020s/>Long-form Video Understanding in the 2020s</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
56text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2024-15-reversifying-neural-networks-efficient-memory-optimization-strategies-for-finetuning-large-models/>Reversifying Neural Networks: Efficient Memory Optimization Strategies for Finetuning Large Models</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
57text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2024-16-towards-more-realistic-continual-learning-at-scale/>Towards More Realistic Continual Learning at Scale</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
58text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2024-06-oppo/>Optimizing Memory Efficiency in Pretrained Model Finetuning</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
59text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2024-17-optimizing-memory-efficiency-in-pretrained-model-finetuning/>Optimizing Memory Efficiency in Pretrained Model Finetuning</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
60text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2024-18-toward-long-form-video-understanding/>Toward Long-form Video Understanding</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
61text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2024-01-pkusz/>Challenges and Advances in Long-form Video Understanding</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
62text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2024-01-13-sustech/>Towards Long-form Video Understanding</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
63text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2023-20-challenges-and-innovation-for-long-form-video-understanding-compute-algorithm-and-data/>Challenges and Innovation for Long-form Video Understanding: Compute, Algorithm, and Data</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
64text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2023-12-sribd/>Research Highlights at IVUL with a Focus on Video Understanding</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
65text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2022-12-chuksz/>Towards Long-form Video Understanding</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
66text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2023-02-rising-start/>A Simple and Effective Approach for Long-form Video Understanding</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
67text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2021-22-detecting-actions-in-videos-via-graph-convolutional-networks/>Detecting Actions in Videos via Graph Convolutional Networks</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
68text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2021-10-sigs/>Let the Computer See the World as Humans</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
69text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2015-10-dalian/>Image/Video Cloud Coding</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
70text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2016-10-msra/>Making a Lighter Encoder: Image/Video Compressive Sensing</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
71text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/talks/2016-04-sony/>Compressive Sensing-based Image/Video Coding</a></li></ul></div></li><li><a class="hb-sidebar-custom-link
72text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/projects/>Projects
73<span data-hb-sidebar-toggle><svg fill="none" viewBox="0 0 24 24" stroke="currentColor" class="h-[18px] min-w-[18px] rounded-sm p-0.5 hover:bg-gray-800/5 dark:hover:bg-gray-100/5"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7" class="origin-center transition-transform rtl:-rotate-180"/></svg></span></a><div class="ltr:pr-0 overflow-hidden"><ul class=hb-sidebar-list><li class="flex flex-col"><a class="hb-sidebar-custom-link
74text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/projects/pandas/>Pandas</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
75text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/projects/pytorch/>PyTorch</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
76text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/projects/scikit/>scikit-learn</a></li></ul></div></li><li><a class="hb-sidebar-custom-link
77text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/>Awards
78<span data-hb-sidebar-toggle><svg fill="none" viewBox="0 0 24 24" stroke="currentColor" class="h-[18px] min-w-[18px] rounded-sm p-0.5 hover:bg-gray-800/5 dark:hover:bg-gray-100/5"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7" class="origin-center transition-transform rtl:-rotate-180"/></svg></span></a><div class="ltr:pr-0 overflow-hidden"><ul class=hb-sidebar-list><li class="flex flex-col"><a class="hb-sidebar-custom-link
79text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2024-01-first-place-ego4d-visual-queries-3d/>First place, Ego4D Visual Queries 3D</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
80text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2024-03-first-place-epic-kitchens-action-detection/>First place, Epic-Kitchens action detection</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
81text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2024-04-first-place-epic-kitchens-action-recognition/>First place, Epic-Kitchens action recognition</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
82text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2024-02-first-place-epic-kitchens-audio-based-interaction-detection/>First place, Epic-Kitchens audio-based interaction detection</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
83text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2023-05-best-paper-award-copy/>Recipient of Grant</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
84text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2023-05-best-paper-award/>Best Paper Award</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
85text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2023-06-first-place-visual-queries-3d-localization-challenge-in-ego4d-workshop/>First place, Visual Queries 3D Localization Challenge in Ego4D Workshop</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
86text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2022-07-first-place-visual-queries-3d-localization-challenge-in-ego4d-workshop/>First place, Visual Queries 3D Localization Challenge in Ego4D Workshop</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
87text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2021-08-outstanding-reviewer/>Outstanding Reviewer</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
88text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2020-09-finalist/>Finalist</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
89text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2020-10-second-place-hacs-temporal-action-localization-challenge/>Second place, HACS Temporal Action Localization Challenge</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
90text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2019-11-finalist/>Finalist</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
91text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2016-12-outstanding-graduate/>Outstanding Graduate</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
92text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2016-13-scholarship-of-out
92standing-talent/>Scholarship of Outstanding Talent</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
93text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2015-14-best-paper-award/>Best Paper Award</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
94text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2012-15-first-prize-qualcomm-innovation-fellowship-contest-qinf/>First Prize, Qualcomm Innovation Fellowship Contest (QInF)</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
95text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2012-16-outstanding-individual-in-the-summer-social-practice/>Outstanding Individual in the Summer Social Practice</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
96text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2010-17-outstanding-graduate-leader/>Outstanding Graduate Leader</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
97text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2008-18-goldman-sachs-global-leaders-award/>Goldman Sachs Global Leaders Award</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
98text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2007-20-first-class-scholarship/>First-Class Scholarship</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
99text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/awards/2007-19-national-scholarship/>National Scholarship</a></li></ul></div></li><li><a class="hb-sidebar-custom-link
100text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/slides/>Slides
101<span data-hb-sidebar-toggle><svg fill="none" viewBox="0 0 24 24" stroke="currentColor" class="h-[18px] min-w-[18px] rounded-sm p-0.5 hover:bg-gray-800/5 dark:hover:bg-gray-100/5"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7" class="origin-center transition-transform rtl:-rotate-180"/></svg></span></a><div class="ltr:pr-0 overflow-hidden"><ul class=hb-sidebar-list><li class="flex flex-col"><a class="hb-sidebar-custom-link
102text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/slides/example/>Example Talk: Recent Work</a></li></ul></div></li><li><a class="hb-sidebar-custom-link
103text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/experience/>Experience</a></li><li><a class="hb-sidebar-custom-link
104text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/>Courses
105<span data-hb-sidebar-toggle><svg fill="none" viewBox="0 0 24 24" stroke="currentColor" class="h-[18px] min-w-[18px] rounded-sm p-0.5 hover:bg-gray-800/5 dark:hover:bg-gray-100/5"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7" class="origin-center transition-transform rtl:-rotate-180"/></svg></span></a><div class="ltr:pr-0 overflow-hidden"><ul class=hb-sidebar-list><li class="flex flex-col"><a class="hb-sidebar-custom-link
106text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/>Hugo Blox
107<span data-hb-sidebar-toggle><svg fill="none" viewBox="0 0 24 24" stroke="currentColor" class="h-[18px] min-w-[18px] rounded-sm p-0.5 hover:bg-gray-800/5 dark:hover:bg-gray-100/5"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7" class="origin-center transition-transform rtl:-rotate-180"/></svg></span></a><div class="ltr:pr-0 overflow-hidden"><ul class=hb-sidebar-list><li class="flex flex-col"><a class="hb-sidebar-custom-link
108text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/getting-started/>Getting Started</a></li><li class="flex flex-col open"><a class="hb-sidebar-custom-link
109text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/guide/>Guide
110<span data-hb-sidebar-toggle><svg fill="none" viewBox="0 0 24 24" stroke="currentColor" class="h-[18px] min-w-[18px] rounded-sm p-0.5 hover:bg-gray-800/5 dark:hover:bg-gray-100/5"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7" class="origin-center transition-transform rtl:-rotate-180"/></svg></span></a><div class="ltr:pr-0 overflow-hidden"><ul class=hb-sidebar-list><li class="flex flex-col"><a class="hb-sidebar-custom-link
111text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/guide/project-structure/>Project Structure</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
112text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/guide/configuration/>Configuration</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
113text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/guide/formatting/>Formatting
114<span data-hb-sidebar-toggle><svg fill="none" viewBox="0 0 24 24" stroke="currentColor" class="h-[18px] min-w-[18px] rounded-sm p-0.5 hover:bg-gray-800/5 dark:hover:bg-gray-100/5"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7" class="origin-center transition-transform rtl:-rotate-180"/></svg></span></a><div class="ltr:pr-0 overflow-hidden"><ul class=hb-sidebar-list><li class="flex flex-col"><a class="hb-sidebar-custom-link
115text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/guide/formatting/media/>Embed Media</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
116text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/guide/formatting/button/>Buttons</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
117text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/guide/formatting/callout/>Callouts</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
118text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/guide/formatting/cards/>Cards</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
119text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/guide/formatting/toggle/>Spoilers</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
120text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/guide/formatting/steps/>Steps</a></li></ul></div></li></ul></div></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
121text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/reference/>Reference
122<span data-hb-sidebar-toggle><svg fill="none" viewBox="0 0 24 24" stroke="currentColor" class="h-[18px] min-w-[18px] rounded-sm p-0.5 hover:bg-gray-800/5 dark:hover:bg-gray-100/5"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7" class="origin-center transition-transform rtl:-rotate-180"/></svg></span></a><div class="ltr:pr-0 overflow-hidden"><ul class=hb-sidebar-list><li class="flex flex-col"><a class="hb-sidebar-custom-link
123text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/reference/customization/>Customization</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
124text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/courses/hugo-blox/reference/i18n/>Internationalization (i18n)</a></li></ul></div></li></ul></div></li></ul></div></li><li><a class="hb-sidebar-custom-link
125text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/news/>News</a></li><li class=open><a class="hb-sidebar-custom-link
126text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/>Publications
127<span data-hb-sidebar-toggle><svg fill="none" viewBox="0 0 24 24" stroke="currentColor" class="h-[18px] min-w-[18px] rounded-sm p-0.5 hover:bg-gray-800/5 dark:hover:bg-gray-100/5"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7" class="origin-center transition-transform rtl:-rotate-180"/></svg></span></a><div class="ltr:pr-0 overflow-hidden"><ul class=hb-sidebar-list><li class="flex flex-col open"><a class="hb-sidebar-custom-link
128sidebar-active-item bg-primary-100 font-semibold text-primary-800 dark:bg-primary-300 dark:text-primary-900" href=/publication/beta_clip/>Beta-CLIP: Text-Conditioned Contrastive Learning for Multi-Granular Vision-Language Alignment</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
129text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/bolt/>BOLT: Boost Large Vision-Language Model Without Training for Long-form Video Understanding</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
130text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/smile-fida/>SMILE: Infusing Spatial and Motion Semantics in Masked Video Learning</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
131text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/osmamba/>OSMamba: Omnidirectional Spectral Mamba with Dual-Domain Prior Generator for Exposure Correction</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
132text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/idm/>
132Invertible Diffusion Models for Compressed Sensing</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
133text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/severe++/>SEVERE++: Evaluating Benchmark Sensitivity in Generalization of Video Representation Learning</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
134text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/maxpool/>Effectiveness of Max-Pooling for Fine-Tuning CLIP on Videos</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
135text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/opentad/>OpenTAD: A Unified Framework and Comprehensive Study of Temporal Action Detection</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
136text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/ego4d-tpami/>Ego4D: Around the World in 3,000 Hours of Egocentric Video</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
137text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/tgt/>Towards Automated Movie Trailer Generation</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
138text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/dr2net/>Dr&lt;sup>2&lt;/sup>Net: Dynamic Reversible Dual-Residual Networks for Memory-Efficient Finetuning</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
139text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/ego-exo4d/>Ego-Exo4D: Understanding Skilled Human Activity from First-and Third-Person Perspectives</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
140text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/adatad/>End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
141text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/re2tal/>Re&lt;sup>2&lt;/sup>TAL: Rewiring Pretrained Video Backbones for Reversible Temporal Action Localization</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
142text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/egoloc/>EgoLoc: Revisiting 3D Object Localization from Egocentric Videos with Visual Queries</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
143text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/freedom/>FreeDoM: Training-Free Energy-Guided Conditional Diffusion Model</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
144text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/lae/>A Unified Continual Learning Framework with General Parameter-Efficient Tuning</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
145text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/lf-vsn/>Large-capacity and Flexible Video Steganography via Invertible Neural Network</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
146text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/etad/>ETAD: Training Action Detection End to End on a Laptop</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
147text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/smile/>Just a Glimpse: Rethinking Temporal Information for Video Continual Learning</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
148text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/owl/>Owl (observe, watch, listen): Localizing actions in egocentric video via audiovisual temporal context</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
149text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/r-dfcil/>R-DFCIL: Relation-Guided Representation Learning for Data-Free Class Incremental Learning</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
150text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/easee/>End-to-End Active Speaker Detection</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
151text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/remote-sensing_wheat-disease/>Evaluation of Diverse Convolutional Neural Networks and Training Strategies for Wheat Leaf Disease Identification with Field-Acquired Photographs</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
152text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/tnas/>When NAS Meets Trees: An Efficient Algorithm for Neural Architecture Search</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
153text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/mad/>MAD: A Scalable Dataset for Language Grounding in Videos from Movie Audio Descriptions</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
154text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/ego4d/>Ego4D: Around the World in 3,000 Hours of Egocentric Video</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
155text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/segtad/>SegTAD: Precise Temporal Action Detection via Semantic Segmentation</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
156text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/vsgn/>Video Self‑Stitching Graph Network for Temporal Action Localization</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
157text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/thumbnet/>
157ThumbNet: One Thumbnail Image Contains All You Need for Recognition</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
158text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/hacs_challenge/>Improve Baseline for Temporal Action Detection: HACS Challenge 2020 Solution of IVUL‑KAUST team</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
159text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/g-tad/>G‑TAD: Sub‑Graph Localization for Temporal Action Detection</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
160text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/optimizationinspired-compact-deep-compressive-sensing/>Optimization‑Inspired Compact Deep Compressive Sensing</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
161text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/youtube8m_challenge/>Logistic Regression is Still Alive and Effective: The 3rd YouTube 8M Challenge Solution of the IVUL‑KAUST team</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
162text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/cream/>CREAM: CNN-REgularized ADMM framework for compressive-sensed image reconstruction</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
163text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/boostnet/>BoostNet: A Structured Deep Recursive Network to Boost Image Deblocking</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
164text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/admm-cnn/>Better and Faster, when ADMM Meets CNN: Compressive-sensed Image Reconstruction</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
165text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/reducing-image-compression-artifacts-by-structural-sparse-representation-and-quantization-constraint-prior/>Reducing Image Compression Artifacts by Structural Sparse Representation and Quantization Constraint Prior</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
166text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/video-compressive-sensing-reconstruction-via-reweighted-residual-sparsity/>Video Compressive Sensing Reconstruction via Reweighted Residual Sparsity</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
167text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/concolor/>CONCOLOR: COnstrained Non-Convex Low-Rank Model for Image Deblocking</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
168text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/dcc16_nonconvex/>Nonconvex Lp Nuclear Norm based ADMM Framework for Compressive Sensing</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
169text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/dcc16_dpcm/>Compressive-Sensed Image Coding via Stripe-based DPCM</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
170text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/a-dual-structured-sparsity-model-for-compressive-sensed-video-reconstruction/>A Dual Structured-Sparsity Model for Compressive-Sensed Video Reconstruction</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
171text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/an-efficient-image-coding-method-based-on-cloud-data/>基于云数据的高效图像编码方法</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
172text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/adaptive-intra-refresh-for-low-delay-error-resilient-video-coding/>Adaptive intra-refresh for low-delay error-resilient video coding</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
173text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/video-compressive-sensing-vi
173a-structured-laplacian-modelling/>Video Compressive Sensing via Structured Laplacian Modelling</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
174text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/icme14_imagecs_multi-hypothesis/>Image Compressive-Sensing Recovery Using Structured Laplacian Sparsity in DCT Domain and Multi-Hypothesis Prediction</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
175text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/sp14_sparsifying-l0/>Image Compressive Sensing Recovery Using Adaptively Learned Sparsifying Basis via L0 Minimization</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
176text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/weakly-supervised-photo-cropping/>Weakly Supervised Photo Cropping</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
177text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/icip13_wavlet/>Wavelet Inpainting Driven Image Compression via Collaborative Sparsity at Low Bit Rates</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
178text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/publication/iscas13_error-concealment/>A Highly Effective Error Concealment Method for Whole Frame Loss</a></li></ul></div></li><li><a class="hb-sidebar-custom-link
179text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/events/>Recent & Upcoming Talks
180<span data-hb-sidebar-toggle><svg fill="none" viewBox="0 0 24 24" stroke="currentColor" class="h-[18px] min-w-[18px] rounded-sm p-0.5 hover:bg-gray-800/5 dark:hover:bg-gray-100/5"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7" class="origin-center transition-transform rtl:-rotate-180"/></svg></span></a><div class="ltr:pr-0 overflow-hidden"><ul class=hb-sidebar-list><li class="flex flex-col"><a class="hb-sidebar-custom-link
181text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/events/example/>Create Beautiful Presentations with Markdown</a></li></ul></div></li><li><a class="hb-sidebar-custom-link
182text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/teaching/>Teaching
183<span data-hb-sidebar-toggle><svg fill="none" viewBox="0 0 24 24" stroke="currentColor" class="h-[18px] min-w-[18px] rounded-sm p-0.5 hover:bg-gray-800/5 dark:hover:bg-gray-100/5"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7" class="origin-center transition-transform rtl:-rotate-180"/></svg></span></a><div class="ltr:pr-0 overflow-hidden"><ul class=hb-sidebar-list><li class="flex flex-col"><a class="hb-sidebar-custom-link
184text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/teaching/machine-learning-cs229/>Machine Learning (CS229), KAUST</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
185text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/teaching/deep-learning-visual-computing-cs323/>Deep Learning for Visual Computing (CS323), KAUST</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
186text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/teaching/video-understanding/>Video understanding, KAUST</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
187text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/teaching/machine-learning-kaust/>Machine learning, KAUST</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
188text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/teaching/deep-learning-kaust/>Deep learning, KAUST</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
189text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/teaching/deep-learning-kaust-bootcamp/>Deep learning, KAUST</a></li><li class="flex flex-col"><a class="hb-sidebar-custom-link
190text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/teaching/deep-learning-kaust-spring-2022/>Deep learning, KAUST</a></li></ul></div></li><li><a class="hb-sidebar-custom-link
191text-gray-500 hover:bg-gray-100 hover:text-gray-900 dark:text-gray-300 dark:hover:bg-primary-800 dark:hover:text-gray-50" href=/contact-success/>Thanks for reaching out</a></li></ul><div class="max-xl:hidden h-0 w-64 shrink-0"></div></div></aside><nav class="hb-toc order-last hidden w-64 shrink-0 xl:block print:hidden px-4" aria-label="table of contents"><div class="hb-scrollbar text-sm [hyphens:auto] sticky top-16 overflow-y-auto pr-4 pt-6 max-h-[calc(100vh-var(--navbar-height)-env(safe-area-inset-bottom))] -mr-4 rtl:-ml-4">
192    
193    
194  </div></nav><article class="w-full break-words flex min-h-[calc(100vh-var(--navbar-height))] min-w-0 justify-center pb-8 pr-[calc(env(safe-area-inset-right)-1.5rem)]"><main class="w-full min-w-0 max-w-6xl px-6 pt-4 md:px-12"><h1 class="mt-2 text-4xl font-bold tracking-tight text-slate-900 dark:text-slate-100" data-pagefind-meta=title data-pagefind-body>Beta-CLIP: Text-Conditioned Contrastive Learning for Multi-Granular Vision-Language Alignment</h1><div class="mt-4 mb-16"><div class="text-gray-500 dark:text-gray-300 text-sm flex items-center flex-wrap gap-y-2"><span class=mr-1>February 21, 2026</span><span class=mx-1>·</span>
195<script>console.log("Author notes detected in single.html, Alpine.js should load")</script>
195<div class="group inline-flex items-center text-current gap-x-1.5 mx-1"><div>Fatimah zohra</div></div><div class="group inline-flex items-center text-current gap-x-1.5 mx-1"><img src=/media/authors/me_hu_2f0cd9e05f21d3cb.webp width=16 height=16 alt="Chen Zhao" class="inline-block h-4 w-4 rounded-full border border-current" loading=lazy><div>Chen Zhao</div></div><span class="relative inline-block ml-1" x-data="{ tooltip: false }"><button @mouseenter="tooltip = true" @mouseleave="tooltip = false" @click="tooltip = !tooltip" class="author-notes text-primary-600 dark:text-primary-400 hover:text-primary-800 dark:hover:text-primary-200 transition-colors cursor-help" data-tooltip=† aria-label=† type=button>
196<svg class="inline-block w-4 h-4" fill="currentColor" viewBox="0 0 20 20"><path fill-rule="evenodd" d="M18 10A8 8 0 112 10a8 8 0 0116 0zm-7-4A1 1 0 119 6a1 1 0 012 0zM9 9a1 1 0 000 2v3a1 1 0 001 1h1a1 1 0 100-2v-3a1 1 0 00-1-1H9z" clip-rule="evenodd"/></svg></button><div x-show=tooltip x-transition:enter="transition ease-out duration-200" x-transition:enter-start="opacity-0 transform scale-95" x-transition:enter-end="opacity-100 transform scale-100" x-transition:leave="transition ease-in duration-150" x-transition:leave-start="opacity-100 transform scale-100" x-transition:leave-end="opacity-0 transform scale-95" @click.away="tooltip = false" class="absolute z-50 bottom-full left-1/2 transform -translate-x-1/2 mb-2 px-3 py-2 text-sm text-white bg-gray-900 dark:bg-gray-700 rounded-lg shadow-lg whitespace-nowrap" x-cloak>†<div class="absolute top-full left-1/2 transform -translate-x-1/2 -mt-1 w-0 h-0 border-l-4 border-r-4 border-t-4 border-transparent border-t-gray-900 dark:border-t-gray-700"></div></div></span><span class=mr-1>,</span><div class="group inline-flex items-center text-current gap-x-1.5 mx-1"><div>Hani itani</div></div><span class=mr-1>,</span><div class="group inline-flex items-center text-current gap-x-1.5 mx-1"><div>Bernard ghanem</div></div><span class=mx-1>·</span>
197<span class=mx-1>0 min read</span></div><div class=mt-3><div><button class="hb-attachment-link hb-attachment-link-large js-cite-clipboard cursor-pointer" type=button data-filename=/publication/beta_clip/cite.bib>
198<svg style="height:1em" class="inline-block" viewBox="0 0 24 24"><path fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.5" d="M15.75 17.25v3.375c0 .621-.504 1.125-1.125 1.125h-9.75A1.125 1.125.0 013.75 20.625V7.875c0-.621.504-1.125 1.125-1.125H6.75a9.06 9.06.0 011.5.124m7.5 10.376h3.375c.621.0 1.125-.504 1.125-1.125V11.25c0-4.46-3.243-8.161-7.5-8.876a9.06 9.06.0 00-1.5-.124H9.375c-.621.0-1.125.504-1.125 1.125v3.5m7.5 10.375H9.375A1.125 1.125.0 018.25 16.125v-9.25m12 6.625v-1.875A3.375 3.375.0 0016.875 8.25h-1.5A1.125 1.125.0 0114.25 7.125v-1.5A3.375 3.375.0 0010.875 2.25H9.75"/></svg>
199<span>Cite</span>
200</button>
201<a class="hb-attachment-link hb-attachment-link-large" href=https://arxiv.org/pdf/2512.12678 target=_blank rel=noopener><svg style="height:1em" class="inline-block" viewBox="0 0 24 24"><path fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.5" d="M19.5 14.25v-2.625A3.375 3.375.0 0016.125 8.25h-1.5A1.125 1.125.0 0113.5 7.125v-1.5A3.375 3.375.0 0010.125 2.25H8.25m0 12.75h7.5m-7.5 3H12M10.5 2.25H5.625c-.621.0-1.125.504-1.125 1.125v17.25c0 .621.504 1.125 1.125 1.125h12.75c.621.0 1.125-.504 1.125-1.125V11.25a9 9 0 00-9-9"/></svg>
202PDF</a></div></div></div><div class="article-header article-container featured-image-wrapper mt-4 mb-16" style=max-width:720px;max-height:339px><div style=position:relative><img srcset="/publication/beta_clip/featured_hu_f1d3c3ab66a3b8e6.webp 320w, /publication/beta_clip/featured_hu_db75ba569b795a2a.webp 480w, /publication/beta_clip/featured_hu_a7b16ea8de2b5fbc.webp 640w, /publication/beta_clip/featured_hu_aae45db504cff282.webp 720w, /publication/beta_clip/featured_hu_7079772b71588c57.webp 720w" sizes="(max-width: 480px) 100vw, (max-width: 640px) 90vw, (max-width: 720px) 80vw, 720px" src=/publication/beta_clip/featured_hu_f1d3c3ab66a3b8e6.webp data-pagefind-meta=image[src] width=720 height=339 alt class=featured-image fetchpriority=high></div></div><div class="max-w-prose grid grid-cols-1 md:grid-cols-[200px_auto] gap-4 my-6"><div class="font-bold text-2xl">
202Abstract</div><div>CLIP achieves strong zero-shot image-text retrieval by aligning global vision and text representations, yet it falls behind on fine-grained tasks even when fine-tuned on long, detailed captions. In this work, we propose beta-CLIP, a multi-granular text-conditioned contrastive learning framework designed to achieve hierarchical alignment between multiple textual granularities-from full captions to sentences and phrases-and their corresponding visual regions. For each level of granularity, beta-CLIP utilizes cross-attention to dynamically pool image patches, producing contextualized visual embeddings. To address the semantic overlap inherent in this hierarchy, we introduce the beta-Contextualized Contrastive Alignment Loss (beta-CAL). This objective parameterizes the trade-off between strict query-specific matching and relaxed intra-image contextualization, supporting both soft Cross-Entropy and hard Binary Cross-Entropy formulations. Through extensive experiments, we demonstrate that beta-CLIP significantly improves dense alignmentv &ndash; achieving 91.8% T2I 92.3% I2T at R@1 on Urban1K and 30.9% on FG-OVD (Hard), setting state-of-the-art among methods trained without hard negatives. beta-CLIP establishes a robust, adaptive baseline for dense vision-language correspondence. The code and models are released at <a href=https://github.com/fzohra/B-CLIP target=_blank rel=noopener>https://github.com/fzohra/B-CLIP</a>.</div><div class="font-bold text-2xl">Type</div><div><a href=/publication_types/1/>1</a></div><div class="font-bold text-2xl">Publication</div><div>arXiv preprint arXiv:2512.12678, 2026</div></div><div class="prose prose-slate lg:prose-xl dark:prose-invert" data-pagefind-body><span data-pagefind-filter=type style=display:none>publication</span></div><time class="mt-12 mb-8 block text-xs text-gray-500 ltr:text-right rtl:text-left dark:text-gray-400" datetime=2026-03-01T14:00:42.000Z><span>Last updated on</span>
203March 1, 2026</time><div class="container mx-auto prose prose-slate lg:prose-xl dark:prose-invert mt-5"><div class="max-w-prose print:hidden"><div class="flex justify-center"><a class="no-underline bg-primary-100 hover:bg-primary-300 text-primary-800 text-xs font-medium mr-2 px-2.5 py-0.5 lg:px-5 lg:py-2 rounded dark:bg-primary-900 dark:hover:bg-primary-700 dark:text-primary-300" href=/tags/deep-learning/>Deep Learning</a>
204<a class="no-underline bg-primary-100 hover:bg-primary-300 text-primary-800 text-xs font-medium mr-2 px-2.5 py-0.5 lg:px-5 lg:py-2 rounded dark:bg-primary-900 dark:hover:bg-primary-700 dark:text-primary-300" href=/tags/clip/>CLIP</a>
205<a class="no-underline bg-primary-100 hover:bg-primary-300 text-primary-800 text-xs font-medium mr-2 px-2.5 py-0.5 lg:px-5 lg:py-2 rounded dark:bg-primary-900 dark:hover:bg-primary-700 dark:text-primary-300" href=/tags/image-retrieval/>Image Retrieval</a>
206<a class="no-underline bg-primary-100 hover:bg-primary-300 text-primary-800 text-xs font-medium mr-2 px-2.5 py-0.5 lg:px-5 lg:py-2 rounded dark:bg-primary-900 dark:hover:bg-primary-700 dark:text-primary-300" href=/tags/fine-grained-computer-vision/>Fine-Grained Computer Vision</a></div><section class="flex flex-row flex-wrap justify-center pt-4 text-xl"><a target=_blank rel=noopener class="m-1 rounded-md bg-neutral-300 p-1.5 text-neutral-700 hover:bg-primary-500 hover:text-neutral-300 dark:bg-neutral-700 dark:text-neutral-300 dark:hover:bg-primary-400 dark:hover:text-neutral-800" href="https://x.com/intent/tweet?url=https%3A%2F%2Fzhao-chen.com%2Fpublication%2Fbeta_clip%2F&amp;text=Beta-CLIP%3A+Text-Conditioned+Contrastive+Learning+for+Multi-Granular+Vision-Language+Alignment" title=X aria-label=X id=share-link-x><svg style="height:1em" viewBox="0 0 512 512"><path fill="currentColor" d="M389.2 48h70.6L305.6 224.2 487 464H345L233.7 318.6 106.5 464H35.8L200.7 275.5 26.8 48H172.4L272.9 180.9 389.2 48zM364.4 421.8h39.1L151.1 88h-42L364.4 421.8z"/></svg>
207</a><a target=_blank rel=noopener class="m-1 rounded-md bg-neutral-300 p-1.5 text-neutral-700 hover:bg-primary-500 hover:text-neutral-300 dark:bg-neutral-700 dark:text-neutral-300 dark:hover:bg-primary-400 dark:hover:text-neutral-800" href="https://www.facebook.com/sharer.php?u=https%3A%2F%2Fzhao-chen.com%2Fpublication%2Fbeta_clip%2F&amp;t=Beta-CLIP%3A+Text-Conditioned+Contrastive+Learning+for+Multi-Granular+Vision-Language+Alignment" title=Facebook aria-label=Facebook i
207d=share-link-facebook><svg style="height:1em" viewBox="0 0 24 24"><path fill="currentColor" d="M22 12c0-5.52-4.48-10-10-10S2 6.48 2 12c0 4.84 3.44 8.87 8 9.8V15H8v-3h2V9.5C10 7.57 11.57 6 13.5 6H16v3h-2c-.55.0-1 .45-1 1v2h3v3h-3v6.95c5.05-.5 9-4.76 9-9.95z"/></svg>
208</a><a target=_blank rel=noopener class="m-1 rounded-md bg-neutral-300 p-1.5 text-neutral-700 hover:bg-primary-500 hover:text-neutral-300 dark:bg-neutral-700 dark:text-neutral-300 dark:hover:bg-primary-400 dark:hover:text-neutral-800" href="mailto:?subject=Beta-CLIP%3A%20Text-Conditioned%20Contrastive%20Learning%20for%20Multi-Granular%20Vision-Language%20Alignment&amp;body=https%3A%2F%2Fzhao-chen.com%2Fpublication%2Fbeta_clip%2F" title=Email aria-label=Email id=share-link-email><svg style="height:1em" viewBox="0 0 24 24"><path fill="none" stroke="currentColor" stroke-linecap="round" stroke-width="1.5" d="M16.5 12a4.5 4.5.0 11-9 0 4.5 4.5.0 019 0zm0 0c0 1.657 1.007 3 2.25 3S21 13.657 21 12a9 9 0 10-2.636 6.364M16.5 12V8.25"/></svg>
209</a><a target=_blank rel=noopener class="m-1 rounded-md bg-neutral-300 p-1.5 text-neutral-700 hover:bg-primary-500 hover:text-neutral-300 dark:bg-neutral-700 dark:text-neutral-300 dark:hover:bg-primary-400 dark:hover:text-neutral-800" href="https://www.linkedin.com/shareArticle?url=https%3A%2F%2Fzhao-chen.com%2Fpublication%2Fbeta_clip%2F&amp;title=Beta-CLIP%3A+Text-Conditioned+Contrastive+Learning+for+Multi-Granular+Vision-Language+Alignment" title=LinkedIn aria-label=LinkedIn id=share-link-linkedin><svg style="height:1em" height="1em" viewBox="0 0 448 512"><path fill="currentColor" d="M416 32H31.9C14.3 32 0 46.5.0 64.3v383.4C0 465.5 14.3 480 31.9 480H416c17.6.0 32-14.5 32-32.3V64.3c0-17.8-14.4-32.3-32-32.3zM135.4 416H69V202.2h66.5V416zm-33.2-243c-21.3.0-38.5-17.3-38.5-38.5S80.9 96 102.2 96c21.2.0 38.5 17.3 38.5 38.5.0 21.3-17.2 38.5-38.5 38.5zm282.1 243h-66.4V312c0-24.8-.5-56.7-34.5-56.7-34.6.0-39.9 27-39.9 54.9V416h-66.4V202.2h63.7v29.2h.9c8.9-16.8 30.6-34.5 62.9-34.5 67.2.0 79.7 44.3 79.7 101.9V416z"/></svg>
210</a><a target=_blank rel=noopener class="m-1 rounded-md bg-neutral-300 p-1.5 text-neutral-700 hover:bg-primary-500 hover:text-neutral-300 dark:bg-neutral-700 dark:text-neutral-300 dark:hover:bg-primary-400 dark:hover:text-neutral-800" href="whatsapp://send?text=Beta-CLIP%3A+Text-Conditioned+Contrastive+Learning+for+Multi-Granular+Vision-Language+Alignment%20https%3A%2F%2Fzhao-chen.com%2Fpublication%2Fbeta_clip%2F" title=WhatsApp aria-label=WhatsApp id=share-link-whatsapp><svg style="height:1em" viewBox="0 0 256 256" fill="currentColor"><path d="m187.58 144.84-32-16a8 8 0 00-8 .5l-14.69 9.8a40.55 40.55.0 01-16-16l9.8-14.69a8 8 0 00.5-8l-16-32A8 8 0 00104 64a40 40 0 00-40 40 88.1 88.1.0 0088 88 40 40 0 0040-40 8 8 0 00-4.42-7.16zM152 176a72.08 72.08.0 01-72-72 24 24 0 0119.29-23.54l11.48 23L101 118a8 8 0 00-.73 7.51 56.47 56.47.0 0030.15 30.15A8 8 0 00138 155l14.61-9.74 23 11.48A24 24 0 01152 176zM128 24A104 104 0 0036.18 176.88l-11.35 34.05a16 16 0 0020.24 20.24l34.05-11.35A104 104 0 10128 24zm0 192a87.87 87.87.0 01-44.06-11.81 8 8 0 00-6.54-.67L40 216l12.47-37.4a8 8 0 00-.66-6.54A88 88 0 11128 216z"/></svg></a></section><div class="flex pt-12 pb-4"><div class=place-self-center><div class="text-[0.6rem] uppercase leading-3 text-neutral-500 dark:text-neutral-400">Authors</div><div class="font-semibold leading-6 text-neutral-800 dark:text-neutral-300"><span class=inline-block><a href=/authors/fatimah-zohra/ class=no-underline>Fatimah zohra</a></span></div><div class="text-2xl sm:text-lg pt-1"></div></div></div><div class="flex pt-12 pb-4"><img class="mr-4 h-24 w-24 rounded-full" srcset="/media/authors/me_hu_64021adeacd1a6c4.webp 96w, /media/authors/me_hu_455101a30e2e1d51.webp 192w, /media/authors/me_hu_12fc467d7861d4ca.webp 288w, /media/authors/me_hu_17f6bebcca15e6e8.webp 384w" sizes=96px src=/media/authors/me_hu_64021adeacd1a6c4.webp width=96 height=96 alt="Chen Zhao" loading=lazy><div class=place-self-center><div class="text-[0.6rem] uppercase leading-3 text-neutral-500 dark:text-neutral-400">Authors</div><div class="font-semibold leading-6 text-neutral-800 dark:text-neutral-300"><span class=inline-block><a href=https://zhao-chen.com/ class=no-underline>Chen Zhao</a></span></div><div class="text-sm font-bold text-neutral-700 dark:text-neutral-300">Professor</div><div class="text-2xl sm:text-lg pt-1"><div class="flex flex-wrap text-neutral-500 dark:text-neutral-300"><a class="pr-2 transition-transform hover:scale-125 hover:text-primary-700 dark:hover:text-primary-400" style=will-change:transform href=mailto:[email protected] aria-label=At-Symbol><svg style="height:1em" viewBox="0 0 24 24">
210<path fill="none" stroke="currentColor" stroke-linecap="round" stroke-width="1.5" d="M16.5 12a4.5 4.5.0 11-9 0 4.5 4.5.0 019 0zm0 0c0 1.657 1.007 3 2.25 3S21 13.657 21 12a9 9 0 10-2.636 6.364M16.5 12V8.25"/></svg></a>
211<a class="pr-2 transition-transform hover:scale-125 hover:text-primary-700 dark:hover:text-primary-400" style=will-change:transform href=https://github.com/coolbay target=_blank rel=noopener rel="me noopener noreferrer" aria-label=Brands/Github><svg style="height:1em" fill="currentColor" viewBox="3 3 18 18"><path d="M12 3C7.0275 3 3 7.12937 3 12.2276c0 4.0833 2.57625 7.5321 6.15374 8.7548C9.60374 21.0631 9.77249 20.7863 9.77249 20.5441 9.77249 20.3249 9.76125 19.5982 9.76125 18.8254 7.5 19.2522 6.915 18.2602 6.735 17.7412 6.63375 17.4759 6.19499 16.6569 5.8125 16.4378 5.4975 16.2647 5.0475 15.838 5.80124 15.8264 6.51 15.8149 7.01625 16.4954 7.18499 16.7723 7.99499 18.1679 9.28875 17.7758 9.80625 17.5335 9.885 16.9337 10.1212 16.53 10.38 16.2993 8.3775 16.0687 6.285 15.2728 6.285 11.7432c0-1.0035.34875-1.834.92249-2.47994C7.1175 9.03257 6.8025 8.08674 7.2975 6.81794c0 0 .753749999999999-.24223 2.47499.94583.72001-.20762 1.48501-.31143 2.25001-.31143C12.7875 7.45234 13.5525 7.55615 14.2725 7.76377c1.7212-1.19959 2.475-.94583 2.475-.94583C17.2424 8.08674 16.9275 9.03257 16.8375 9.26326 17.4113 9.9092 17.76 10.7281 17.76 11.7432c0 3.5411-2.1037 4.3255-4.1063 4.5561C13.98 16.5877 14.2613 17.1414 14.2613 18.0065 14.2613 19.2407 14.25 20.2326 14.25 20.5441 14.25 20.7863 14.4188 21.0746 14.8688 20.9824 16.6554 20.364 18.2079 19.1866 19.3078 17.6162c1.0999-1.5705 1.6917-3.4551 1.6922-5.3886C21 7.12937 16.9725 3 12 3z"/></svg></a>
212<a class="pr-2 transition-transform hover:scale-125 hover:text-primary-700 dark:hover:text-primary-400" style=will-change:transform href="https://scholar.google.com/citations?user=dUWdX5EAAAAJ&amp;hl=en" target=_blank rel=noopener rel="me noopener noreferrer" aria-label=Academicons/Google-Scholar><svg style="height:1em" viewBox="0 0 512 512"><path fill="currentColor" d="M343.759 106.662V79.43L363.524 64h-213.89L20.476 176.274h85.656a82.339 82.339.0 00-.219 6.225c0 20.845 7.22 38.087 21.672 51.861 14.453 13.797 32.252 20.648 53.327 20.648 4.923.0 9.75-.368 14.438-1.024-2.907 6.5-4.374 12.523-4.374 18.142.0 9.875 4.499 20.43 13.467 31.642-39.234 2.67-68.061 9.732-86.437 21.163-10.531 6.5-19 14.704-25.39 24.531-6.391 9.9-9.578 20.515-9.578 31.962.0 9.648 2.062 18.336 6.219 26.062 4.156 7.726 9.578 14.07 16.312 18.984 6.718 4.968 14.469 9.101 23.219 12.469 8.734 3.344 17.406 5.718 26.061 7.062A167.052 167.052.0 00180.555 448c13.469.0 26.953-1.734 40.547-5.187 13.562-3.485 26.28-8.642 38.171-15.493 11.86-6.805 21.515-16.086 28.922-27.718 7.39-11.68 11.094-24.805 11.094-39.336.0-11.016-2.25-21.039-6.75-30.14-4.468-9.073-9.938-16.542-16.452-22.345-6.501-5.813-13-11.155-19.516-15.968-6.5-4.845-12-9.75-16.468-14.813-4.485-5.046-6.735-10.054-6.735-14.984.0-4.921 1.734-9.672 5.216-14.265 3.455-4.61 7.674-9.048 12.61-13.306 4.937-4.25 9.875-8.968 14.796-14.133 4.922-5.147 9.141-11.827 12.61-20.008 3.485-8.18 5.203-17.445 5.203-27.757.0-13.453-2.547-24.46-7.547-33.314-.594-1.022-1.218-1.803-1.875-3.022l56.907-46.672v17.119c-7.393.93-6.624 5.345-6.624 10.635V245.96c0 5.958 4.875 10.834 10.834 10.834h3.989c5.958.0 10.833-4.875 10.833-10.834V117.293c0-5.277.778-9.688-6.561-10.63zm-107.36 222.48c1.14.75 3.704 2.78 7.718 6.038 4.05 3.243 6.797 5.695 8.266 7.414a443.553 443.553.0 016.376 7.547c2.813 3.375 4.718 6.304 5.718 8.734 1 2.477 2.016 5.461 3.047 8.946a38.27 38.27.0 011.485 10.562c0 17.048-6.564 29.68-19.656 37.859-13.125 8.18-28.767 12.274-46.938 12.274-9.187.0-18.203-1.093-27.063-3.196-8.843-2.116-17.311-5.336-25.39-9.601-8.078-4.258-14.577-10.204-19.5-17.797-4.938-7.64-7.407-16.415-7.407-26.25.0-10.32 2.797-19.29 8.422-26.906 5.594-7.625 12.938-13.391 22.032-17.315 9.063-3.946 18.25-6.742 27.562-8.398a157.865 157.865.0 0128.438-2.555c4.47.0 7.936.25 10.405.696.455.219 3.032 2.07 7.735 5.563 4.704 3.462 7.625 5.595 8.75 6.384zm-3.359-100.579c-7.406 8.86-17.734 13.288-30.953 13.288-11.86.0-22.298-4.764-31.266-14.312-9-9.523-15.422-20.328-19.344-32.43-3.937-12.109-5.906-23.984-5.906-35.648.0-13.694 3.596-25.352 10.781-34.976 7.187-9.65 17.5-14.485 30.938-14.485 11.875.0 22.374 5.038 31.437 15.157 9.094 10.085 15.61 21.413 19.517 33.968 3.922 12.54 5.873 24.53 5.873 35.984.0 13.446-3.702 24.61-11.076 33.454z"/></svg></a></div></div></div></div><div class="flex pt-12 pb-4">
212<div class=place-self-center><div class="text-[0.6rem] uppercase leading-3 text-neutral-500 dark:text-neutral-400">Authors</div><div class="font-semibold leading-6 text-neutral-800 dark:text-neutral-300"><span class=inline-block><a href=/authors/hani-itani/ class=no-underline>Hani itani</a></span></div><div class="text-2xl sm:text-lg pt-1"></div></div></div><div class="flex pt-12 pb-4"><div class=place-self-center><div class="text-[0.6rem] uppercase leading-3 text-neutral-500 dark:text-neutral-400">Authors</div><div class="font-semibold leading-6 text-neutral-800 dark:text-neutral-300"><span class=inline-block><a href=/authors/bernard-ghanem/ class=no-underline>Bernard ghanem</a></span></div><div class="text-2xl sm:text-lg pt-1"></div></div></div><div class="pt-1 no-prose w-full"><hr class="border-dotted border-neutral-300 dark:border-neutral-600"><div class="flex flex-col md:flex-row flex-nowrap justify-between gap-5 pt-2"><div></div><div><a class="group flex text-right no-underline" href=/publication/bolt/><span class="flex flex-col"><span class="mt-[0.1rem] leading-6 group-hover:underline group-hover:decoration-primary-500">BOLT: Boost Large Vision-Language Model Without Training for Long-form Video Understanding</span>
213<span class="mt-[0.1rem] text-xs text-neutral-500 dark:text-neutral-400">October 29, 2025
214</span></span><span class="mt-[-0.3rem] ms-2 text-neutral-700 transition-transform group-hover:-translate-x-[2px] group-hover:text-primary-600 dark:text-neutral dark:group-hover:text-primary-400"><span class=ltr:inline>&rarr;</span></span></a></div></div></div></div></div></main></article></div></div><div class=page-footer><footer class="container mx-auto flex flex-col justify-items-center text-sm leading-6 mt-24 mb-4 text-slate-700 dark:text-slate-200"><p class="powered-by text-center">© All rights reserved by Chen Zhao, 2026</p></footer></div><div id=hb-notification-container class="fixed top-20 right-4 z-[9999] pointer-events-none" aria-live=polite aria-atomic=true></div></body></html>

Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.