Research map / 3D vision / Depth and stereo
ORCaS: Unsupervised Depth Completion via Occluded Region Completion as Supervision
Hyoungseob Park, Runjian Chen, Patrick Rim, Dong Lao, Alex Wong
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add ORCaS loss (feature supervision of occluded regions) on top of 2D-to-3D broadcast and 3D warping (ablation) | ORCaS · VOID1500 | MAE: 40.52 -> 30.90 | Improved |
| add 3D warping without ORCaS loss (ablation) | base model · VOID1500 | MAE: 35.31 -> 52.60 | Hurt |
| add 2D-to-3D broadcasting to base network (ablation) | KBNet with transformer · VOID1500 | MAE: 35.31 -> 33.56 | Improved |
| add 3D broadcast to warping plus ORCaS loss (ablation) | ORCaS · VOID1500 | MAE: 36.37 -> 30.90 | Improved |
| use feature supervision instead of depth supervision for occluded region (ablation) | ORCaS · VOID1500 | MAE: 38.81 -> 30.90 | Improved |
| add feature supervision versus no supervision (ablation) | ORCaS · VOID1500 | MAE: 33.56 -> 30.90 | Improved |
| increase depth planes from 2 to 8 (ablation) | ORCaS · VOID1500 | MAE: 32.73 -> 30.90 | Improved |
| increase context pooling size from (2,2) to (4,4) (ablation) | ORCaS · VOID1500 | MAE: 32.86 -> 30.90 | Improved |
| increase context pooling size from (4,4) to (8,8) (ablation) | ORCaS · VOID1500 | MAE: 30.90 -> 31.36 | Hurt |
| mask moving objects during training (ablation) | ORCaS · VOID1500 | MAE: 30.90 -> 30.84 | No clear effect |
| replace AugUndo with ORCaS for zero-shot transfer | AugUndo · ScanNet (zero-shot from VOID1500) | MAE: 82.53 -> 68.86 | Improved |
| replace AugUndo with ORCaS under sparse input | AugUndo · VOID150 | MAE: 117.93 -> 81.89 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DesNet — DesNet: Decomposed Scale-Consistent Network for Unsupervised Depth Completion (AAAI 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.88)
- PacGDC: Label-Efficient Generalizable Depth Completion with Projection Ambiguity and Consistency (ICCV 2025 · similarity 0.88)
- DeCoTR: Enhancing Depth Completion with 2D and 3D Attentions (CVPR 2024 · similarity 0.88)
- VoxDet: Rethinking 3D Semantic Scene Completion as Dense Object Detection (NeurIPS 2025 · similarity 0.87)
- Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior (AAAI 2025 · similarity 0.87)
- Boosting Self-Supervision for Single-View Scene Completion via Knowledge Distillation (CVPR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
