Research map / 3D vision / 3D generation
Geometry-Aware Recurrent Neural Networks for Active Visual Recognition
Ricson Cheng, Ziyan Wang, Katerina Fragkiadaki
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 |
|---|---|---|---|
| use the learned Active policy instead of 1-step greedy | visual reconstruction · single object, view-2 | IoU increase over single-view reconstruction (%): 29.1% -> 27.7%±3.23% | Hurt |
| use the learned Active policy instead of 1-step greedy | visual reconstruction · single object, view-3 | IoU increase over single-view reconstruction (%): 32.7% -> 34.1%±2.08% | Improved |
| use the learned Active policy instead of 1-step greedy | visual reconstruction · single object, view-4 | IoU increase over single-view reconstruction (%): 33.3% -> 36.6%±1.17% | Improved |
| use the learned Active policy instead of 1-step greedy | visual reconstruction · multi-objects, view-2 | IoU increase over single-view reconstruction (%): 36.2% -> 31.3%±2.26% | Hurt |
| use the learned Active policy instead of 1-step greedy | visual reconstruction · multi-objects, view-3 | IoU increase over single-view reconstruction (%): 40.4% -> 39.0%±1.99% | Hurt |
| use the learned Active policy instead of 1-step greedy | visual reconstruction · multi-objects, view-4 | IoU increase over single-view reconstruction (%): 40.4% -> 41.8%±2.36% | Improved |
| use the learned Active policy instead of the best sampled trajectory | visual reconstruction · single object, view-4 | IoU increase over single-view reconstruction (%): 38.1% -> 36.6%±1.17% | Hurt |
| use the learned Active policy instead of the best sampled trajectory | visual reconstruction · multi-objects, view-4 | IoU increase over single-view reconstruction (%): 44.7% -> 41.8%±2.36% | Hurt |
| add a fourth view | car reconstruction · car reconstruction, view-3 to view-4 | reconstruction IoU: 0.86 -> 0.86 | No clear effect |
| replace simple REINFORCE with more elaborate actor-critic methods | view selection policy training · view selection | view selection performance | No clear effect |
Closest papers by meaning
- Learning Spatial Common Sense With Geometry-Aware Recurrent Networks (CVPR 2019 · similarity 0.92)
- 3D-R2N2: A Unified Approach for Single and Multi-view 3D Object Reconstruction (ECCV 2016 · similarity 0.88)
- Scene Representation Networks: Continuous 3D-Structure-Aware Neural Scene Representations (NeurIPS 2019 · similarity 0.87)
- AA-RMVSNet: Adaptive Aggregation Recurrent Multi-View Stereo Network (ICCV 2021 · similarity 0.87)
- Veritatem Dies Aperit - Temporally Consistent Depth Prediction Enabled by a Multi-Task Geometric and Semantic Scene Understanding Approach (CVPR 2019 · similarity 0.86)
- FrameVGGT: Coherence-Preserving Memory for Bounded Streaming Geometry (NeurIPS 2026 · similarity 0.85)
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.
