Research map / 3D vision / Depth and stereo
Where to Explore Next? ExHistCNN for History-aware Autonomous 3D Exploration
Yiming Wang, Alessio Del Bue
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 utility map to depth input (CNN2D vs CNNDepth) (ablation) | ExHistCNN · SUNCG direction classification | avg F1: 0.445 -> 0.632 | Improved |
| partition utility map into four directions (CNN4D vs CNNUtility) (ablation) | ExHistCNN · SUNCG direction classification | avg F1: 0.624 -> 0.635 | Improved |
| add partitioned utility maps to depth (CNN5D vs CNN2D) (ablation) | ExHistCNN · SUNCG direction classification | avg F1: 0.632 -> 0.642 | Improved |
| scale depth to match utility map FoV (CNN2DScaled vs CNN2D) (ablation) | ExHistCNN · SUNCG direction classification | avg F1: 0.632 -> 0.595 | Hurt |
| use pretrained ResNet101 features with MLP instead of CNN trained from scratch (MLP2D vs CNN2D) (ablation) | ExHistCNN · SUNCG direction classification | avg F1: 0.632 -> 0.610 | Hurt |
| use pretrained ResNet101 features with MLP for 5D input (MLP5D vs CNN5D) (ablation) | ExHistCNN · SUNCG direction classification | avg F1: 0.642 -> 0.654 | Improved |
| look two steps ahead instead of one in oracle (OracleDepth2 vs OracleDepth) (ablation) | Oracle NBV · SUNCG synthetic rooms exploration | coverage ratio | Improved |
| look three steps ahead instead of two in oracle (ablation) | Oracle NBV · SUNCG synthetic rooms exploration | coverage ratio | No clear effect |
| use non-overlapping triangular partition instead of rectangular overlapping partition (ablation) | BaseGain · SUNCG synthetic rooms exploration | coverage ratio | Improved |
| add depth to partitioned utility maps (ablation) | CNN4D · SUNCG exploration | coverage ratio | Mixed |
| transfer learned strategies from synthetic to real Matterport3D scans | CNN2D, CNN5D · Matterport3D exploration | coverage ratio | Mixed |
| use CNN2D rather than MLP2D for the NBV direction | ExHistCNN · SUNCG exploration | coverage ratio | Improved |
Closest papers by meaning
- NExplore: Exploration with Neural Fields for Autonomous Scene Reconstruction (TPAMI 2026 · similarity 0.86)
- Move to Understand a 3D Scene: Bridging Visual Grounding and Exploration for Efficient and Versatile Embodied Navigation (ICCV 2025 · similarity 0.85)
- Learn-to-Score: Efficient 3D Scene Exploration by Predicting View Utility (ECCV 2018 · similarity 0.85)
- GenNBV: Generalizable Next-Best-View Policy for Active 3D Reconstruction (CVPR 2024 · similarity 0.84)
- RoomTour3D: Geometry-Aware Video-Instruction Tuning for Embodied Navigation (CVPR 2025 · similarity 0.84)
- Learning with a Mole: Transferable latent spatial representations for navigation without reconstruction (ICLR 2024 · 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.
