Research map / Vision-language and multimodal / Robot action and navigation
PRISM: Learning Realistic Depth via Physics-Grounded Noise Disentanglement with Semantic-Geometric Collaboration
Xiujian Liang, Jiacheng Liu, Mingyang Sun, Qichen He, Anda Cheng, Cewu Lu, Jianhua Sun
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 |
|---|---|---|---|
| train policies on PRISM-synthesized depth instead of clean simulation depth | depth-aware policies · Real-world manipulation, 6 tasks | average success rate (%): 38.9 -> 92.1 | Improved |
| replace Stable-Sim2Real with PRISM noise synthesis | depth-aware policies · Real-world manipulation, 6 tasks | average success rate (%): 89.1 -> 92.1 | Improved |
| replace Stable-Sim2Real with PRISM noise synthesis | depth synthesis · ByteCameraDepth | sensing invalidation IoU: 0.876 -> 0.952 | Improved |
| replace Stable-Sim2Real with PRISM noise synthesis | depth synthesis · NYU-Depth-v2 zero-shot | sensing invalidation IoU: 0.756 -> 0.923 | Improved |
| add Semantic-Physics Reasoner (VFM priors) (ablation) | PRISM · ByteCameraDepth non-Lambertian surfaces | performance | Improved |
| add Bimodal Noise Disentangler (ablation) | PRISM · ByteCameraDepth | overall MAE | Improved |
| use 3D VFM (MoGe2, Depth Anything V2) instead of DINOv2 or ResNet as prior (ablation) | PRISM · ByteCameraDepth ablation | Robot SR | Improved |
| add positive-weighted multi-scale BCE over standard BCE (ablation) | H-PPS · ByteCameraDepth | invalidation recall | Improved |
| add Dice loss (ablation) | H-PPS · ByteCameraDepth | hole compactness | Improved |
| use dynamic decaying hard mining (PP-DHM) instead of static mining (ablation) | H-PPS · ByteCameraDepth | Boundary-IoU: 8.8% (12.0% relative) | Improved |
Closest papers by meaning
- PhyScene3D: Physically Consistent 3D Interactive Tabletop Scene Generation (ICML 2026 · similarity 0.86)
- PRISM3D: Probabilistic Refinement and Robust Initialization for Physically Consistent Scene Modeling under Extreme Motion Blur (ECCV 2026 · similarity 0.86)
- Beyond Static Vision: Scene Dynamic Field Unlocks Intuitive Physics Understanding in Multi-modal Large Language Models (ICLR 2026 · similarity 0.86)
- PRISM: Programming Interactive Scenes from Monocular Images for Embodied Simulation (NeurIPS 2026 · similarity 0.84)
- Understanding while Exploring: Semantics-driven Active Mapping (NeurIPS 2025 · similarity 0.84)
- GEOPHYS: The Geometry of Physical Plausibility (NeurIPS 2026 · similarity 0.84)
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.
