Research map / 3D vision / 3D generation
Deep Supervision With Shape Concepts for Occlusion-Aware 3D Object Parsing
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 sequential supervision for pose, visibility, 3D keypoints, and 2D keypoints (ablation) | DISCO · KITTI-3D | PCK (α=0.1), 2D All: 80.6 -> 85.0 | Helped |
| add visibility supervision (ablation) | DISCO · KITTI-3D | PCK (α=0.1), 2D All: 80.7 -> 85.0 | Helped |
| add 3D keypoint supervision (ablation) | DISCO · KITTI-3D | PCK (α=0.1), 2D All: 83.4 -> 85.0 | Helped |
| add pose supervision (ablation) | DISCO · KITTI-3D | PCK (α=0.1), 2D All: 83.4 -> 85.0 | Helped |
| switch the 3D keypoint and visibility supervision order (ablation) | DISCO · KITTI-3D | PCK (α=0.1), 2D All: 85.0 -> 75.8 | Hurt |
| reverse the full order of intermediate supervision (ablation) | DISCO · KITTI-3D | PCK (α=0.1), 2D All: 85.0 -> 25.4 | Hurt |
| move all supervision to the last convolutional layer (ablation) | DISCO · KITTI-3D | PCK (α=0.1), 2D All: 85.0 -> 80.6 | Hurt |
| replace shape-concept supervision with 2D-only supervision (ablation) | DISCO · KITTI-3D | PCK (α=0.1), 2D All: 85.0 -> 37.5 | Hurt |
| use VGG-style pooling and stacked fully connected layers (ablation) | DISCO · KITTI-3D | PCK (α=0.1), 2D All: 85.0 -> 69.0 | Hurt |
| train on fully visible cars only instead of including simulated occlusion and truncation (ablation) | DISCO · KITTI-3D | 2D keypoint accuracy: decrease by 1.3%; decrease by 24.9%; decrease by 15.9% | Hurt |
| use DISCO rather than VKps | 2D keypoint localization · PASCAL VOC car, Occluded | PCK (α=0.1): 62.8 -> 59.0 | Hurt |
| use DISCO rather than 3D-INN | 3D structure prediction · IKEA sofa | average recall: 88.0 -> 83.4 | Hurt |
Closest papers by meaning
- 3D-RCNN: Instance-Level 3D Object Reconstruction via Render-and-Compare (CVPR 2018 · similarity 0.88)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.88)
- Occlusion-aware Text-Image-Point Cloud Pretraining for Open-World 3D Object Recognition (CVPR 2025 · similarity 0.87)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.87)
- OA-CNNs: Omni-Adaptive Sparse CNNs for 3D Semantic Segmentation (CVPR 2024 · similarity 0.87)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.87)
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.