If you're tired of slow 3D scene completion pipelines, we built RapidLiDAR to hit 10 Hz in real-time. Drop your questions below!</p>\n","updatedAt":"2026-08-20T19:40:25.723Z","author":{"_id":"689f7f336563f6bd1b51817b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/iMtnKGn4hZCbm6hWq7lxs.png","fullname":"Azhar Hussian","name":"Azhar88","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9207643866539001},"editors":["Azhar88"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/iMtnKGn4hZCbm6hWq7lxs.png"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.16490","authors":[{"_id":"6a871a3789e517cbfd75daea","user":{"_id":"689f7f336563f6bd1b51817b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/iMtnKGn4hZCbm6hWq7lxs.png","isPro":false,"fullname":"Azhar Hussian","user":"Azhar88","type":"user","name":"Azhar88"},"name":"Azhar Hussian","status":"claimed_verified","statusLastChangedAt":"2026-08-20T16:45:04.587Z","hidden":false},{"_id":"6a871a3789e517cbfd75daeb","name":"Martin Vossiek","hidden":false},{"_id":"6a871a3789e517cbfd75daec","name":"Vasileios Belagiannis","hidden":false}],"publishedAt":"2026-08-17T00:00:00.000Z","submittedOnDailyAt":"2026-08-20T00:00:00.000Z","title":"Towards Real-Time and Adaptable LiDAR Scene Completion","submittedOnDailyBy":{"_id":"689f7f336563f6bd1b51817b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/iMtnKGn4hZCbm6hWq7lxs.png","isPro":false,"fullname":"Azhar Hussian","user":"Azhar88","type":"user","name":"Azhar88"},"summary":"LiDAR scene completion is a key component of 3D perception in autonomous driving, where the scene must be completed in real time to be usable in downstream tasks. Existing approaches typically follow an initialize-and-refine paradigm, in which a coarse initialization of the scene is first constructed, then refined into complete 3D geometry. Generative models are slower because they iteratively refine random Gaussian noise into the scene, while non-generative methods perturb the partial scene with a fixed noise scale, which limits coverage of large gaps and occluded regions and requires manual recalibration for each new sensor configuration. We present RapidLiDAR, a LiDAR scene completion method that treats the initialization itself as a learned, data-driven component. We propose an adaptive initialization module that predicts a spatially varying displacement for each partial input point, expanding the partial observations into a coarse scene initialization adapted to the local geometry, without requiring manual noise tuning. To refine this coarse initialization into a complete and coherent scene, we additionally propose a multi-scale reconstruction module that further refines point positions by querying multi-scale 3D voxel and 2D BEV feature maps constructed from the input scan. By replacing point-neighborhood operators such as farthest point sampling and k-nearest neighbor search with voxel- and BEV-based feature extraction, our architecture is faster and can handle different input resolutions by design. Experiments on SemanticKITTI and KITTI-360 show that our method achieves completion performance on par with the state of the art while completing a full scene in 0.1 seconds, which is 2.3 times faster than the fastest prior method. This matches the 10 Hz acquisition rate of typical automotive LiDAR sensors, taking a step toward real-time LiDAR scene completion.","upvotes":2,"discussionId":"6a871a3789e517cbfd75daed","githubRepo":"https://github.com/AzharSindhi/RapidLiDAR","githubRepoAddedBy":"user","ai_summary":"RapidLiDAR learns adaptive spatial displacements to initialize LiDAR scenes and refines them via multi-scale voxel and BEV features for real-time completion.","ai_keywords":["LiDAR scene completion","adaptive initialization","spatially varying displacement","multi-scale reconstruction","3D voxel features","2D BEV feature maps","farthest point sampling","k-nearest neighbor search"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":1,"organization":{"_id":"61dea3edc201b96c5db3bedc","name":"fau","fullname":"FAU Erlangen-Nürnberg","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6404403bad54665351d42ee2/tnh0_WIjEr_OGOPcMzLQE.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64771adf03fe88eff54d08eb","avatarUrl":"/avatars/f78cb16e6b1a8a0931b81cf4446af0b6.svg","isPro":false,"fullname":"Azhar Hussian","user":"AzharHussian","type":"user"},{"_id":"689f7f336563f6bd1b51817b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/iMtnKGn4hZCbm6hWq7lxs.png","isPro":false,"fullname":"Azhar Hussian","user":"Azhar88","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"61dea3edc201b96c5db3bedc","name":"fau","fullname":"FAU Erlangen-Nürnberg","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6404403bad54665351d42ee2/tnh0_WIjEr_OGOPcMzLQE.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.16490.md","query":{}}">
Towards Real-Time and Adaptable LiDAR Scene Completion
Abstract
RapidLiDAR learns adaptive spatial displacements to initialize LiDAR scenes and refines them via multi-scale voxel and BEV features for real-time completion.
LiDAR scene completion is a key component of 3D perception in autonomous driving, where the scene must be completed in real time to be usable in downstream tasks. Existing approaches typically follow an initialize-and-refine paradigm, in which a coarse initialization of the scene is first constructed, then refined into complete 3D geometry. Generative models are slower because they iteratively refine random Gaussian noise into the scene, while non-generative methods perturb the partial scene with a fixed noise scale, which limits coverage of large gaps and occluded regions and requires manual recalibration for each new sensor configuration. We present RapidLiDAR, a LiDAR scene completion method that treats the initialization itself as a learned, data-driven component. We propose an adaptive initialization module that predicts a spatially varying displacement for each partial input point, expanding the partial observations into a coarse scene initialization adapted to the local geometry, without requiring manual noise tuning. To refine this coarse initialization into a complete and coherent scene, we additionally propose a multi-scale reconstruction module that further refines point positions by querying multi-scale 3D voxel and 2D BEV feature maps constructed from the input scan. By replacing point-neighborhood operators such as farthest point sampling and k-nearest neighbor search with voxel- and BEV-based feature extraction, our architecture is faster and can handle different input resolutions by design. Experiments on SemanticKITTI and KITTI-360 show that our method achieves completion performance on par with the state of the art while completing a full scene in 0.1 seconds, which is 2.3 times faster than the fastest prior method. This matches the 10 Hz acquisition rate of typical automotive LiDAR sensors, taking a step toward real-time LiDAR scene completion.
Community
If you're tired of slow 3D scene completion pipelines, we built RapidLiDAR to hit 10 Hz in real-time. Drop your questions below!
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2608.16490 in a model README.md to link it from this page.
Cite arxiv.org/abs/2608.16490 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2608.16490 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.