Research map / 3D vision / 3D generation
3D-DLP: Self-supervised 3D Object-centric Scene Representation Learning
Ellina Zhang, Madhavan Iyengar, Amir Zadeh, Chuan Li, David Held, Deepak Pathak, Tal Daniel
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 K-means keypoint proposals instead of spatial softmax on raw voxels (ablation) | 3D-DLP-VC · MimicGen stack (RGB voxels ablation) | Masked PSNR: 13.52 -> 20.15 | Improved |
| add K-means keypoint proposals instead of spatial softmax on learned heatmaps (ablation) | 3D-DLP-VC · MimicGen stack (RGB voxels ablation) | Masked PSNR: 15.06 -> 20.15 | Improved |
| add chroma loss (ablation) | 3D-DLP-VC · MimicGen stack (RGB voxels ablation) | Masked PSNR: 19.37 -> 20.15 | Improved |
| add chroma loss (ablation) | 3D-DLP-VC · MimicGen stack (RGB voxels ablation) | IoU: 0.785 -> 0.806 | Improved |
| increase particles from 8 to 24 (ablation) | 3D-DLP-VC · MimicGen | Masked PSNR: 21.30 -> 24.41 | Improved |
| increase particles from 16 to 24 (ablation) | 3D-DLP-VC · MimicGen | Masked PSNR: 22.48 -> 24.41 | Improved |
| increase particles from 24 to 40 (ablation) | 3D-DLP-VC · MimicGen | Masked PSNR: 24.41 -> 23.93 | No clear effect |
| use object-centric 3D-DLP-VC instead of VAE | reconstruction · MimicGen (RGB voxels) | Masked PSNR: 4.35 -> 24.41 | Improved |
| use 3D-DLP-VC instead of deterministic AE on IoU | reconstruction · RGB-D-SD-v2 (RGB voxels) | IoU: 0.771 -> 0.731 | Hurt |
| use 3D particles instead of 2D-DLP single-view | EC-Diffuser policy · MimicGen 12 tasks | mean success rate (%): 30.8 -> 48.1 | Improved |
| use 3D particles instead of 2D-DLP multi-view | EC-Diffuser policy · MimicGen 12 tasks | mean success rate (%): 34.1 -> 48.1 | Improved |
| use 3D-DLP instead of EquiDiff voxel-only | policy · MimicGen 12 tasks | mean success rate (%): 47.3 -> 48.1 | Improved |
Closest papers by meaning
- Unsupervised Image Representation Learning with Deep Latent Particles (ICML 2022 · similarity 0.89)
- Masked Scene Modeling: Narrowing the Gap Between Supervised and Self-Supervised Learning in 3D Scene Understanding (CVPR 2025 · similarity 0.87)
- Common3D: Self-Supervised Learning of 3D Morphable Models for Common Objects in Neural Feature Space (CVPR 2025 · similarity 0.87)
- Object-Centric Representation Learning for Enhanced 3D Semantic Scene Graph Prediction (NeurIPS 2025 · similarity 0.86)
- Sonata: Self-Supervised Learning of Reliable Point Representations (CVPR 2025 · similarity 0.86)
- Object-X: Learning to Reconstruct Multi-Modal 3D Object Representations (NeurIPS 2025 · 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.
