Research map / Detection and segmentation / 3D detection and driving
Monocular Differentiable Rendering for Self-Supervised 3D Object Detection
Deniz Beker, Hiroharu Kato, Mihai Adrian Morariu, Takahiro Ando, Toru Matsuoka, Wadim Kehl, Adrien Gaidon
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 the depth map loss to the reconstruction loss (ablation) | MonoDR · KITTI 3D val, reconstruction loss ablation (AP_R40 IoU 0.7) | 3D AP Easy: 4.82 -> 12.50 | Improved |
| Add 2D losses to a depth-only objective (ablation) | MonoDR · KITTI 3D val, reconstruction loss ablation (AP_R40 IoU 0.7) | 3D AP Easy: 0.00 -> 12.50 | Improved |
| Add confidence weighting (silhouette reconstruction and box protrusion) (ablation) | MonoDR · KITTI 3D val, confidence weighting ablation (AP_R40 IoU 0.7) | 3D AP Easy: 10.74 -> 12.50 | Improved |
| Add silhouette reconstruction weighting to confidence (ablation) | MonoDR · KITTI 3D val, confidence weighting ablation (AP_R40 IoU 0.7) | 3D AP Easy: 11.19 -> 12.50 | Improved |
| Add box protrusion weighting to confidence (ablation) | MonoDR · KITTI 3D val, confidence weighting ablation (AP_R40 IoU 0.7) | 3D AP Easy: 12.43 -> 12.50 | No clear effect |
| Use Mask R-CNN X152 instead of Panoptic R101-FPN for detection and segmentation (ablation) | MonoDR · KITTI 3D val, segmentation network ablation (AP_R40 IoU 0.7) | 3D AP Easy: 9.26 -> 12.50 | Improved |
| Optimize object shapes instead of random shapes (ablation) | MonoDR · KITTI 3D val, shape ablation (AP_R40 IoU 0.7) | 3D AP Easy: 9.10 -> 12.50 | Improved |
| Optimize object shapes instead of cuboids (ablation) | MonoDR · KITTI 3D val, shape ablation (AP_R40 IoU 0.7) | 3D AP Easy: 0.30 -> 12.50 | Improved |
| Self-supervised MonoDR instead of supervised detectors (Easy cases) | MonoDR · KITTI 3D val, car, AP_R40 IoU 0.7 | 3D AP Easy: 11.06 (MonoDIS) / 14.28 (MoVI-3D) vs 12.50 | Mixed |
| Use monocular depth plus rendering without LiDAR instead of SDFLabel with LiDAR | MonoDR · KITTI 3D val, car, AP_R40 IoU 0.7 | 3D AP Easy: 1.23 -> 12.50 | Improved |
Closest papers by meaning
- Objects Are Different: Flexible Monocular 3D Object Detection (CVPR 2021 · similarity 0.89)
- VSRD: Instance-Aware Volumetric Silhouette Rendering for Weakly Supervised 3D Object Detection (CVPR 2024 · similarity 0.89)
- MonoRUn: Monocular 3D Object Detection by Reconstruction and Uncertainty Propagation (CVPR 2021 · similarity 0.88)
- MonoMAE: Enhancing Monocular 3D Detection through Depth-Aware Masked Autoencoders (NeurIPS 2024 · similarity 0.87)
- MonoDGP: Monocular 3D Object Detection with Decoupled-Query and Geometry-Error Priors (CVPR 2025 · similarity 0.87)
- Weakly Supervised Monocular 3D Detection with a Single-View Image (CVPR 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.
