90% to 99% of per-instance performance across 7 diverse benchmarks.\n* 🏅 **Training-Free:** Surpasses fully-supervised methods requiring complete mask annotations for training.\n* ⚡ Fast, interactive speed on a single GPU with negligible memory overhead.","html":"<p>🔥 <strong>Early Accepted to MICCAI 2026 (Top 9%)</strong></p>\n<p><strong>TL;DR:</strong> We introduce <strong>Chain-of-Prompts (CoP)</strong>, a training-free framework that shifts the interaction cost for cell instance segmentation from <em>per-instance O(N)</em> to <em>per-type O(T)</em> using foundation models like SAM 3.</p>\n<p><strong>The Problem:</strong> While interactive foundation models generalize well, clicking hundreds to thousands of individual cells in digital pathology is highly unscalable.<br><strong>The Solution:</strong> CoP requires only <strong>one click per cell type</strong>. It leverages the natural clustering within SAM's frozen image encoder to recursively propagate this single click to all same-type instances across the image.</p>\n<p><strong>Key Highlights:</strong></p>\n<ul>\n<li>🖱️ <strong>Cost Efficiency:</strong> Over 97% reduction in annotation cost (clicks).</li>\n<li>🎯 <strong>High Performance:</strong> Retains &gt;90% to 99% of per-instance performance across 7 diverse benchmarks.</li>\n<li>🏅 <strong>Training-Free:</strong> Surpasses fully-supervised methods requiring complete mask annotations for training.</li>\n<li>⚡ Fast, interactive speed on a single GPU with negligible memory overhead.</li>\n</ul>\n","updatedAt":"2026-06-01T12:20:31.168Z","author":{"_id":"631dd3d27beada304659b6dc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/631dd3d27beada304659b6dc/g6h7sFtf9nWO0QDtSSv9a.jpeg","fullname":"Sanghyun Jo","name":"shjo-april","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.763381838798523},"editors":["shjo-april"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/631dd3d27beada304659b6dc/g6h7sFtf9nWO0QDtSSv9a.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.29429","authors":[{"_id":"6a1c251f808ddbc3c7d4326e","user":{"_id":"631dd3d27beada304659b6dc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/631dd3d27beada304659b6dc/g6h7sFtf9nWO0QDtSSv9a.jpeg","isPro":false,"fullname":"Sanghyun Jo","user":"shjo-april","type":"user","name":"shjo-april"},"name":"Sanghyun Jo","status":"claimed_verified","statusLastChangedAt":"2026-06-01T09:34:14.644Z","hidden":false},{"_id":"6a1c251f808ddbc3c7d4326f","name":"Seo Jin Lee","hidden":false},{"_id":"6a1c251f808ddbc3c7d43270","name":"Seohyung Hong","hidden":false},{"_id":"6a1c251f808ddbc3c7d43271","name":"Yoorim Gang","hidden":false},{"_id":"6a1c251f808ddbc3c7d43272","name":"Hyeongsub Kim","hidden":false},{"_id":"6a1c251f808ddbc3c7d43273","name":"Hyungseok Seo","hidden":false},{"_id":"6a1c251f808ddbc3c7d43274","name":"Kyungsu Kim","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/631dd3d27beada304659b6dc/uOIiPqkQyyBXdTQMIXg-p.gif"],"publishedAt":"2026-05-28T00:00:00.000Z","submittedOnDailyAt":"2026-06-01T00:00:00.000Z","title":"One Click per Cell Type Suffices: Training-free Group Interaction for Cell Instance Segmentation","submittedOnDailyBy":{"_id":"631dd3d27beada304659b6dc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/631dd3d27beada304659b6dc/g6h7sFtf9nWO0QDtSSv9a.jpeg","isPro":false,"fullname":"Sanghyun Jo","user":"shjo-april","type":"user","name":"shjo-april"},"summary":"Cell instance segmentation models trained on cell-specific datasets suffer severe performance drops on out-of-distribution cell types, while interactive foundation models overcome this through per-instance prompting at a cost that is prohibitively expensive for histopathology images containing hundreds to thousands of densely packed instances. We introduce Group Prompting, a new paradigm that shifts interactive segmentation from per-instance O(N) to per-type O(T), where a single click per cell type suffices to segment all instances of that type. Our key observation is that the frozen image encoder of the Segment Anything Model (SAM) already clusters same-type cells in its feature space before any prompt is given. Exploiting this property, we propose Chain-of-Prompts (CoP), a training-free framework that recursively expands a single user click by (1) identifying reliable same-type locations through non-parametric gating of multi-scale encoder features, and (2) selecting the most spatially distant reliable point as the next prompt to maximize coverage. On three cell-type-annotated benchmarks, CoP with one click per type retains over 90% of per-instance performance and surpasses fully-supervised methods without any additional training. On four morphologically homogeneous benchmarks, a single click retains over 99%. Project Page: https://shjo-april.github.io/Chain-of-Prompts/","upvotes":1,"discussionId":"6a1c2520808ddbc3c7d43275","projectPage":"https://shjo-april.github.io/Chain-of-Prompts/","githubRepo":"https://github.com/shjo-april/Chain-of-Prompts","githubRepoAddedBy":"user","ai_summary":"Group Prompting enables efficient cell instance segmentation by leveraging per-type prompting through a training-free framework that uses multi-scale encoder features and recursive prompt expansion.","ai_keywords":["cell instance segmentation","interactive foundation models","per-instance prompting","per-type prompting","Segment Anything Model","frozen image encoder","non-parametric gating","multi-scale encoder features","Chain-of-Prompts","recursive prompt expansion"],"githubStars":3,"organization":{"_id":"66d54dc8033492801db2bf5a","name":"SeoulNatlUniv","fullname":"Seoul National University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/659ccc9d18897eb6594e897f/_-0BM-1UyM-d-lRiahFnf.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"661ab1f1fa3b144a381fa454","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/661ab1f1fa3b144a381fa454/IlpZBb9NCjo7ntFwMIH53.png","isPro":false,"fullname":"Urro","user":"urroxyz","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"66d54dc8033492801db2bf5a","name":"SeoulNatlUniv","fullname":"Seoul National University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/659ccc9d18897eb6594e897f/_-0BM-1UyM-d-lRiahFnf.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.29429.md"}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2605.29429 in a model README.md to link it from this page.

Cite arxiv.org/abs/2605.29429 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2605.29429 in a Space README.md to link it from this page.

Add this paper to a collection to link it from this page.