Towards Learning to Complete Anything in Lidar
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910912823164928 |
|---|---|
| author | Takmaz, Ayca Saltori, Cristiano Peri, Neehar Meinhardt, Tim de Lutio, Riccardo Leal-Taixé, Laura Ošep, Aljoša |
| author_facet | Takmaz, Ayca Saltori, Cristiano Peri, Neehar Meinhardt, Tim de Lutio, Riccardo Leal-Taixé, Laura Ošep, Aljoša |
| contents | We propose CAL (Complete Anything in Lidar) for Lidar-based shape-completion in-the-wild. This is closely related to Lidar-based semantic/panoptic scene completion. However, contemporary methods can only complete and recognize objects from a closed vocabulary labeled in existing Lidar datasets. Different to that, our zero-shot approach leverages the temporal context from multi-modal sensor sequences to mine object shapes and semantic features of observed objects. These are then distilled into a Lidar-only instance-level completion and recognition model. Although we only mine partial shape completions, we find that our distilled model learns to infer full object shapes from multiple such partial observations across the dataset. We show that our model can be prompted on standard benchmarks for Semantic and Panoptic Scene Completion, localize objects as (amodal) 3D bounding boxes, and recognize objects beyond fixed class vocabularies. Our project page is https://research.nvidia.com/labs/dvl/projects/complete-anything-lidar |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_12264 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Towards Learning to Complete Anything in Lidar Takmaz, Ayca Saltori, Cristiano Peri, Neehar Meinhardt, Tim de Lutio, Riccardo Leal-Taixé, Laura Ošep, Aljoša Computer Vision and Pattern Recognition We propose CAL (Complete Anything in Lidar) for Lidar-based shape-completion in-the-wild. This is closely related to Lidar-based semantic/panoptic scene completion. However, contemporary methods can only complete and recognize objects from a closed vocabulary labeled in existing Lidar datasets. Different to that, our zero-shot approach leverages the temporal context from multi-modal sensor sequences to mine object shapes and semantic features of observed objects. These are then distilled into a Lidar-only instance-level completion and recognition model. Although we only mine partial shape completions, we find that our distilled model learns to infer full object shapes from multiple such partial observations across the dataset. We show that our model can be prompted on standard benchmarks for Semantic and Panoptic Scene Completion, localize objects as (amodal) 3D bounding boxes, and recognize objects beyond fixed class vocabularies. Our project page is https://research.nvidia.com/labs/dvl/projects/complete-anything-lidar |
| title | Towards Learning to Complete Anything in Lidar |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2504.12264 |