Research map / 3D vision / NeRF and radiance fields
DM-NeRF: 3D Scene Geometry Decomposition and Manipulation from 2D Images
Read the paperCheck 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 3D empty-space losses (l3d_empty + l3d_obj) (ablation) | DM-NeRF · DM-SR manipulation: Translation | AP@0.90: 87.26 -> 89.33 | Helped |
| add 3D empty-space losses (l3d_empty + l3d_obj) (ablation) | DM-NeRF · DM-SR manipulation: Joint | AP@0.90: 77.36 -> 81.70 | Helped |
| add visual-occlusion handling step in inverse query (ablation) | DM-NeRF · DM-SR manipulation: Scale | AP@0.90: 76.33 -> 86.05 | Helped |
| add visual-occlusion handling step in inverse query (ablation) | DM-NeRF · DM-SR manipulation: Translation | AP@0.90: 86.93 -> 89.33 | Helped |
| add 3D empty-space losses (delta d 0.05) (ablation) | DM-NeRF · ScanNet decomposition | AP@0.75: 92.29 -> 95.60 | Helped |
| add 3D empty-space losses (delta d 0.05) (ablation) | DM-NeRF · Replica decomposition | AP@0.75: 77.76 -> 81.03 | Helped |
| add 3D empty-space losses (delta d 0.05) (ablation) | DM-NeRF · DM-SR decomposition | AP@0.75: 99.77 -> 99.80 | No clear effect |
| vary delta d among 0.025/0.05/0.10 m (ablation) | DM-NeRF · scene decomposition | AP@0.75 | No clear effect |
| learn object codes in 3D (DM-NeRF) instead of per-image 2D segmentation (Mask-RCNN) | Mask-RCNN · DM-SR with 80% noisy labels | AP@0.75: 3.88 -> 74.08 | Helped |
| learn object codes in 3D instead of fine-tuned Mask-RCNN | Mask-RCNN · ScanNet decomposition | AP@0.75: 86.25 -> 95.60 | Helped |
| learn object codes in 3D instead of fine-tuned Mask-RCNN | Mask-RCNN · Replica Office_4 | AP@0.75: 74.76 -> 70.33 | Hurt |
| fine-tune pretrained Mask-RCNN per scene vs train separately from scratch (ablation) | Mask-RCNN · ScanNet | AP@0.75: 85.95 -> 86.25 | Helped |
Closest papers by meaning
- Total-Decom: Decomposed 3D Scene Reconstruction with Minimal Interaction (CVPR 2024 · similarity 0.88)
- DynaVol: Unsupervised Learning for Dynamic Scenes through Object-Centric Voxelization (ICLR 2024 · similarity 0.87)
- Decompositional Neural Scene Reconstruction with Generative Diffusion Prior (CVPR 2025 · similarity 0.87)
- NeRFDeformer: NeRF Transformation from a Single View via 3D Scene Flows (CVPR 2024 · similarity 0.87)
- Language-driven Object Fusion into Neural Radiance Fields with Pose-Conditioned Dataset Updates (CVPR 2024 · similarity 0.87)
- In-N-Out: Lifting 2D Diffusion Prior for 3D Object Removal via Tuning-Free Latents Alignment (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.