Research map / 3D vision / 3D generation
Feed-Forward SceneDINO for Unsupervised Semantic Scene Completion
Aleksandar Jevtić, Christoph Reich, Felix Wimbauer, Oliver Hahn, Christian Rupprecht, Stefan Roth, Daniel Cremers
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 the downsampler for high-resolution target features (vs bilinear upsampling + augmentation) (ablation) | SceneDINO · SSCBench-KITTI-360 (51.2m) | mIoU: 6.82 -> 8.00 | Improved |
| add the feature smoothness loss (ablation) | SceneDINO · SSCBench-KITTI-360 (51.2m) | mIoU: 6.83 -> 8.00 | Improved |
| add positional-encoding decomposition (ablation) | SceneDINO · SSCBench-KITTI-360 (51.2m) | mIoU: 7.26 -> 8.00 | Improved |
| use estimated ORB-SLAM3 poses instead of KITTI-360 poses (ablation) | SceneDINO · SSCBench-KITTI-360 (51.2m) | mIoU: 8.00 -> 7.88 | No clear effect |
| switch target features from DINO to DINOv2 (ablation) | SceneDINO · SSCBench-KITTI-360 (51.2m) | mIoU: 8.00 -> 9.08 | Improved |
| add 3D feature distillation (vs clustering raw features) (ablation) | SceneDINO · SSCBench-KITTI-360 (51.2m) | mIoU: 6.39 -> 8.00 | Improved |
| add kNN-correlation loss (ablation) | SceneDINO · SSCBench-KITTI-360 (51.2m) | mIoU: 6.65 -> 8.00 | Improved |
| add neighborhood sampling around center points (ablation) | SceneDINO · SSCBench-KITTI-360 (51.2m) | mIoU: 7.03 -> 8.00 | Improved |
| use 3D sampling instead of 5-crop sampling (ablation) | SceneDINO · SSCBench-KITTI-360 (51.2m) | mIoU: 7.53 -> 8.00 | Improved |
| lift DINO features into a 3D feature field with distillation vs S4C + STEGO baseline | SceneDINO · SSCBench-KITTI-360 (51.2m) | mIoU: 6.60 -> 8.00 | Improved |
| add 2D-supervised linear probing instead of unsupervised probing | SceneDINO (DINO) · SSCBench-KITTI-360 (51.2m) | mIoU: 8.00 -> 9.34 | Improved |
| use FiT3D 3D-refined features instead of DINO in STEGO | STEGO · KITTI-360 2D segmentation | mIoU: 23.57 -> 22.29 | Hurt |
Closest papers by meaning
- 3D Semantic Scene Completion: A Survey (IJCV 2022 · similarity 0.93)
- Disentangling Instance and Scene Contexts for 3D Semantic Scene Completion (ICCV 2025 · similarity 0.91)
- Bi-SSC: Geometric-Semantic Bidirectional Fusion for Camera-based 3D Semantic Scene Completion (CVPR 2024 · similarity 0.90)
- Memory-Augmented Re-Completion for 3D Semantic Scene Completion (AAAI 2025 · similarity 0.89)
- IPFormer: Visual 3D Panoptic Scene Completion with Context-Adaptive Instance Proposals (NeurIPS 2025 · similarity 0.89)
- H2GFormer: Horizontal-to-Global Voxel Transformer for 3D Semantic Scene Completion (AAAI 2024 · similarity 0.89)
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.
