38.54 mIoU with one correction step. Three commands reproduce that number in about six minutes on one H100 once the base predictions are local.\n- We also report 39.2% mIoU / 59.0 IoU_cmpl with four correction steps and 8-view D4 TTA. Codabench shows each team's best entry, so that is the row visible there, but it sits outside the single-sample setting and the +2.1 pp margin above is not read off it. The per-panel IoU numbers in the figures posted with this submission are from that four-step configuration, not from the N=1 headline.\n- The correction pass costs 107.2 ms: 9.33 FPS for the pass, 3.23 FPS end to end.\n\nThe paper also covers SGSC, from-noise generative completion at 30.5% val mIoU.\n\n- Code: https://github.com/BillyChern/GSSC-S2D2\n- Project page: https://shichen.world/GSSC-project-page/\n- Checkpoints: https://huggingface.co/Stone-Chern/GSSC-S2D2-checkpoints\n- PS3 synthetic scenes: https://huggingface.co/datasets/Stone-Chern/PS3-SemanticKITTI\n- Base predictions: https://huggingface.co/datasets/Stone-Chern/GSSC-S2D2-datasets\n- Leaderboard: https://www.codabench.org/competitions/13814/#/results-tab","html":"<p><strong>GSSC replaces a frozen scene-completion model's argmax with one cheap correction step.</strong> The base network (SCPNet) is left untouched; we run a single discrete-diffusion refinement pass on the probability simplex over its output. No retraining of the base, no distillation, no test-time adaptation.</p>\n<ul>\n<li><strong>38.8% mIoU on the SemanticKITTI hidden test set</strong>, single sweep, N=1, no test-time augmentation. To our knowledge the best causal, single-sweep, single-sample result on the leaderboard, and +2.1 pp over the previous best published score under that restriction (SCPNet, 36.7).</li>\n<li><strong>+2.36 pp on val seq 08</strong>: 36.17 -&gt; 38.54 mIoU with one correction step. Three commands reproduce that number in about six minutes on one H100 once the base predictions are local.</li>\n<li>We also report 39.2% mIoU / 59.0 IoU_cmpl with four correction steps and 8-view D4 TTA. Codabench shows each team's best entry, so that is the row visible there, but it sits outside the single-sample setting and the +2.1 pp margin above is not read off it. The per-panel IoU numbers in the figures posted with this submission are from that four-step configuration, not from the N=1 headline.</li>\n<li>The correction pass costs 107.2 ms: 9.33 FPS for the pass, 3.23 FPS end to end.</li>\n</ul>\n<p>The paper also covers SGSC, from-noise generative completion at 30.5% val mIoU.</p>\n<ul>\n<li>Code: <a href=\"https://github.com/BillyChern/GSSC-S2D2\" rel=\"nofollow\">https://github.com/BillyChern/GSSC-S2D2</a></li>\n<li>Project page: <a href=\"https://shichen.world/GSSC-project-page/\" rel=\"nofollow\">https://shichen.world/GSSC-project-page/</a></li>\n<li>Checkpoints: <a href=\"https://huggingface.co/Stone-Chern/GSSC-S2D2-checkpoints\">https://huggingface.co/Stone-Chern/GSSC-S2D2-checkpoints</a></li>\n<li>PS3 synthetic scenes: <a href=\"https://huggingface.co/datasets/Stone-Chern/PS3-SemanticKITTI\">https://huggingface.co/datasets/Stone-Chern/PS3-SemanticKITTI</a></li>\n<li>Base predictions: <a href=\"https://huggingface.co/datasets/Stone-Chern/GSSC-S2D2-datasets\">https://huggingface.co/datasets/Stone-Chern/GSSC-S2D2-datasets</a></li>\n<li>Leaderboard: <a href=\"https://www.codabench.org/competitions/13814/#/results-tab\" rel=\"nofollow\">https://www.codabench.org/competitions/13814/#/results-tab</a></li>\n</ul>\n","updatedAt":"2026-08-31T16:04:56.042Z","author":{"_id":"670a3300dc8f500b50aaa65c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/670a3300dc8f500b50aaa65c/LW3n691D_cGEADv8VcRjO.jpeg","fullname":"Shi Chen","name":"Stone-Chern","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8157506585121155},"editors":["Stone-Chern"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/670a3300dc8f500b50aaa65c/LW3n691D_cGEADv8VcRjO.jpeg"],"reactions":[{"reaction":"🔥","users":["Stone-Chern"],"count":1}],"isReport":false}},{"id":"6a95a67d4df814820429be83","author":{"_id":"670a3300dc8f500b50aaa65c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/670a3300dc8f500b50aaa65c/LW3n691D_cGEADv8VcRjO.jpeg","fullname":"Shi Chen","name":"Stone-Chern","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false},"createdAt":"2026-08-31T16:06:21.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Author here, happy to answer questions.\n\nTwo things the abstract compresses that are worth stating plainly:\n\nThe 38.8 headline is one correction step, one sweep, no ensembling. That\nrestriction is what the +2.1 pp is indexed on. Codabench displays each team's\nbest entry, so the row visible there is the four-step + 8-view D4 TTA one at\n39.2 / 59.0 IoU_cmpl, which sits outside that restriction.\n\nThe correction pass costs 107 ms, 3.23 FPS end to end. That is an incremental\npass, not a deployable rate.\n\nCode, weights and the PS3 corpus are released. If something does not reproduce,\nopen an issue on the repo and I will look at it.\n","html":"<p>Author here, happy to answer questions.</p>\n<p>Two things the abstract compresses that are worth stating plainly:</p>\n<p>The 38.8 headline is one correction step, one sweep, no ensembling. That<br>restriction is what the +2.1 pp is indexed on. Codabench displays each team's<br>best entry, so the row visible there is the four-step + 8-view D4 TTA one at<br>39.2 / 59.0 IoU_cmpl, which sits outside that restriction.</p>\n<p>The correction pass costs 107 ms, 3.23 FPS end to end. That is an incremental<br>pass, not a deployable rate.</p>\n<p>Code, weights and the PS3 corpus are released. If something does not reproduce,<br>open an issue on the repo and I will look at it.</p>\n","updatedAt":"2026-08-31T16:06:21.656Z","author":{"_id":"670a3300dc8f500b50aaa65c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/670a3300dc8f500b50aaa65c/LW3n691D_cGEADv8VcRjO.jpeg","fullname":"Shi Chen","name":"Stone-Chern","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9356247782707214},"editors":["Stone-Chern"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/670a3300dc8f500b50aaa65c/LW3n691D_cGEADv8VcRjO.jpeg"],"reactions":[{"reaction":"❤️","users":["Stone-Chern"],"count":1}],"isReport":false}},{"id":"6a962b639c5bedfc6886e320","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":379,"isUserFollowing":false},"createdAt":"2026-09-01T01:33:23.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [FPSGen: Flexible Point Cloud Scene Generation with BEV-Supported Transport Flows](https://huggingface.co/papers/2607.26645) (2026)\n* [BIMScript: Material-Aware Structured Scene Programs for BIM Ingestion](https://huggingface.co/papers/2608.21447) (2026)\n* [Iterate or Widen? When Test-Time Refinement Helps LiDAR Scene Completion: A Controlled Study of Evidence Geometry, Training Coverage, and Compute](https://huggingface.co/papers/2608.06014) (2026)\n* [Training-Free Open-Vocabulary 3D Point-Cloud Segmentation on the Generalized Few-Shot Benchmark](https://huggingface.co/papers/2607.15331) (2026)\n* [GeoBridge: Decoupled Semantic Conditioning for Generative Image Geolocalization](https://huggingface.co/papers/2608.11838) (2026)\n* [SparseOcc++: Geometry-Aware Sparse Latent Representation for Semantic Occupancy Prediction](https://huggingface.co/papers/2607.04732) (2026)\n* [Towards Real-Time and Adaptable LiDAR Scene Completion](https://huggingface.co/papers/2608.16490) (2026)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"<p>This is an automated message from the <a href=\"https://huggingface.co/librarian-bots\">Librarian Bot</a>. I found the following papers similar to this paper. </p>\n<p>The following papers were recommended by the Semantic Scholar API </p>\n<ul>\n<li><a href=\"https://huggingface.co/papers/2607.26645\">FPSGen: Flexible Point Cloud Scene Generation with BEV-Supported Transport Flows</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2608.21447\">BIMScript: Material-Aware Structured Scene Programs for BIM Ingestion</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2608.06014\">Iterate or Widen? When Test-Time Refinement Helps LiDAR Scene Completion: A Controlled Study of Evidence Geometry, Training Coverage, and Compute</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2607.15331\">Training-Free Open-Vocabulary 3D Point-Cloud Segmentation on the Generalized Few-Shot Benchmark</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2608.11838\">GeoBridge: Decoupled Semantic Conditioning for Generative Image Geolocalization</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2607.04732\">SparseOcc++: Geometry-Aware Sparse Latent Representation for Semantic Occupancy Prediction</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2608.16490\">Towards Real-Time and Adaptable LiDAR Scene Completion</a> (2026)</li>\n</ul>\n<p> Please give a thumbs up to this comment if you found it helpful!</p>\n<p> If you want recommendations for any Paper on Hugging Face checkout <a href=\"https://huggingface.co/spaces/librarian-bots/recommend_similar_papers\">this</a> Space</p>\n<p> You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: <code>@librarian-bot recommend</code></p>\n","updatedAt":"2026-09-01T01:33:23.648Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":379,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7064926028251648},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.26737","authors":[{"_id":"6a90e614a64059bab69c356c","user":{"_id":"670a3300dc8f500b50aaa65c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/670a3300dc8f500b50aaa65c/LW3n691D_cGEADv8VcRjO.jpeg","isPro":true,"fullname":"Shi Chen","user":"Stone-Chern","type":"user","name":"Stone-Chern"},"name":"Shi Chen","status":"claimed_verified","statusLastChangedAt":"2026-08-28T08:45:04.591Z","hidden":false},{"_id":"6a90e614a64059bab69c356d","name":"Weifeng Ge","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/670a3300dc8f500b50aaa65c/YRDTuY4uGIEmxAtp-NQ5c.png","https://cdn-uploads.huggingface.co/production/uploads/670a3300dc8f500b50aaa65c/w4iuK-kAt34dultRd9Veq.png"],"publishedAt":"2026-08-27T00:00:00.000Z","submittedOnDailyAt":"2026-08-31T00:00:00.000Z","title":"Generative Semantic Scene Completion","submittedOnDailyBy":{"_id":"670a3300dc8f500b50aaa65c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/670a3300dc8f500b50aaa65c/LW3n691D_cGEADv8VcRjO.jpeg","isPro":true,"fullname":"Shi Chen","user":"Stone-Chern","type":"user","name":"Stone-Chern"},"summary":"Outdoor LiDAR semantic scene completion (SSC) recovers a dense semantic voxel grid from a scan observing 1% of the target volume, under class imbalance beyond 7,000x. We recast SSC as generative semantic scene completion (GSSC): a single discrete-diffusion formulation in three roles. First, paired sparse-dense scene synthesis (PS^3) generates matched sparse LiDAR observations with their dense semantic completions, addressing the long tail at its source and yielding the PS^3-SemanticKITTI corpus we train on alongside SemanticKITTI. Second, semantic-guided generative scene completion (SGSC) generates the scene from noise with multinomial discrete diffusion, conditioned on the sparse scan through a bird's-eye-view semantic map and a sparse 3D feature stream. Third, the same framework instead refines an existing completion in one flow-matching step: structured source discrete diffusion (S^2D^2). S^2D^2 improves the mIoU of SGSC's own output and every external SSC base tested, without base retraining or test-time adaptation. On the strongest base, one step without test-time augmentation reaches 38.8% mIoU on the SemanticKITTI hidden test. To our knowledge that is the best causal, single-sweep, single-sample result on that leaderboard, +2.1 pp over the previous best published score under the same restriction. Four correction steps with eight-view test-time augmentation reach 39.2%, outside that restriction.","upvotes":2,"discussionId":"6a90e614a64059bab69c356e","projectPage":"https://shichen.world/GSSC-project-page/","githubRepo":"https://github.com/BillyChern/GSSC-S2D2","githubRepoAddedBy":"user","ai_summary":"Generative discrete diffusion recasts outdoor LiDAR semantic scene completion via paired sparse-dense synthesis, semantic-guided generation, and structured refinement to achieve state-of-the-art results.","ai_keywords":["semantic scene completion","discrete diffusion","multinomial diffusion","flow matching","bird's-eye-view semantic map","sparse 3D feature stream","paired sparse-dense scene synthesis","structured source discrete diffusion"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":3},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"670a3300dc8f500b50aaa65c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/670a3300dc8f500b50aaa65c/LW3n691D_cGEADv8VcRjO.jpeg","isPro":true,"fullname":"Shi Chen","user":"Stone-Chern","type":"user"},{"_id":"6a2da6c8ca070ee12c6e396c","avatarUrl":"/avatars/0355287dcabaa67dbc7f0b10b87451f9.svg","isPro":false,"fullname":"Joe Mama","user":"JoeMama123123123","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.26737.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2608.26737 in a Space README.md to link it from this page.