Research map / Video / Video segmentation and tracking
DeepVideoMVS: Multi-View Stereo on Video With Recurrent Spatio-Temporal Fusion
Check 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 spatio-temporal fusion to the pair network | DeepVideoMVS · ScanNet, 7-Scenes, RGB-D V2, TUM RGB-D, Augmented ICL-NUIM | abs-inv: 0.0468 -> 0.0381; 0.0653 -> 0.0537; 0.0667 -> 0.0566; 0.0736 -> 0.0551; 0.0402 -> 0.0297 | Helped |
| Warp the previous hidden state into the current viewpoint (ablation) | DeepVideoMVS fusion · All test sequences | abs: 0.1495 -> 0.1318 | Helped |
| Replace ConvLSTM with ConvGRU (ablation) | DeepVideoMVS fusion · ScanNet validation split | abs-inv: 0.0340 -> 0.0359 | Hurt |
| Use ELU instead of tanh in the cell (ablation) | ConvLSTM fusion cell · ScanNet validation split | abs: 0.1321 -> 0.1267 | Helped |
| Add cell-state layer normalization to ELU cell (ablation) | ConvLSTM fusion cell · ScanNet validation split | abs: 0.1267 -> 0.1236 | Helped |
| Remove layer normalization before sigmoid gates (ablation) | ConvLSTM fusion cell · ScanNet validation split | abs: 0.1236 -> 0.1206 | Helped |
| Finetune the cell using the test-time predicted-depth warping strategy (ablation) | DeepVideoMVS fusion · ScanNet validation split | abs-inv: 0.0346 -> 0.0338 | Helped |
| Remove low-resolution feature-pyramid skip connections (ablation) | Fusion Network · ScanNet validation split | abs-inv: 0.0340 -> 0.0355 | Hurt |
| Increase plane-sweep hypotheses from M=64 to M=88 (ablation) | Pair Network · ScanNet validation split | abs-inv: 0.0427 -> 0.0422 | Helped |
| Use fusion instead of increasing plane-sweep hypotheses to M=88 (ablation) | DeepVideoMVS · ScanNet validation split | abs-inv: 0.0422 -> 0.0340 | Helped |
Closest papers by meaning
- DeepMVS: Learning Multi-View Stereopsis (CVPR 2018 · similarity 0.90)
- RIAV-MVS: Recurrent-Indexing an Asymmetric Volume for Multi-View Stereo (CVPR 2023 · similarity 0.89)
- MVS2D: Efficient Multi-View Stereo via Attention-Driven 2D Convolutions (CVPR 2022 · similarity 0.89)
- A Global Depth-Range-Free Multi-View Stereo Transformer Network with Pose Embedding (NeurIPS 2024 · similarity 0.88)
- RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · similarity 0.87)
- MVSFormer++: Revealing the Devil in Transformer's Details for Multi-View Stereo (ICLR 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.