Research map / 3D vision / Depth and stereo
TransMVSNet: Global Context-Aware Multi-View Stereo Network With Transformers
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 focal loss to the CasMVSNet-based baseline (ablation) | TransMVSNet baseline · DTU evaluation set | Overall (mm): 0.345 -> 0.339 | Helped |
| Add FMT with intra- and inter-attention (ablation) | TransMVSNet baseline with focal loss · DTU evaluation set | Overall (mm): 0.339 -> 0.323 | Helped |
| Add the transformed feature pathway (ablation) | TransMVSNet with FMT · DTU evaluation set | Overall (mm): 0.323 -> 0.315 | Helped |
| Add the ARF module (ablation) | TransMVSNet with FMT and transformed feature pathway · DTU evaluation set | Overall (mm): 0.315 -> 0.305 | Helped |
| Use 3 input views rather than 5 (ablation) | TransMVSNet · DTU evaluation set | Overall (mm): 0.305 -> 0.316 | Hurt |
| Use 7 or 9 input views rather than 5 (ablation) | TransMVSNet · DTU evaluation set | Overall (mm): 0.305 -> 0.311; 0.305 -> 0.317 | Mixed |
| Reduce input resolution from 864×1152 to 512×640 (ablation) | TransMVSNet · DTU evaluation set | Overall (mm): 0.305 -> 0.361 | Hurt |
| Set focal-loss gamma to 0.5, 1, or 2 instead of 0 (ablation) | TransMVSNet · DTU evaluation set | Overall (mm): 0.305 -> 0.312; 0.305 -> 0.310; 0.305 -> 0.314 | Mixed |
| Use gamma=2 rather than gamma=0 on BlendedMVS (ablation) | TransMVSNet focal loss · BlendedMVS validation | EPE: 0.80 -> 0.73 | Helped |
| Change the number of attention heads from 8 to 1, 2, 4, or 16 (ablation) | TransMVSNet FMT · DTU evaluation set | Overall (mm): 0.305 -> 0.309; 0.305 -> 0.306; 0.305 -> 0.306; 0.305 -> 0.306 | Mixed |
| Use 2 rather than 4 attention blocks (ablation) | TransMVSNet FMT · DTU evaluation set | Overall (mm): 0.305 -> 0.312 | Hurt |
| Use 6 rather than 4 attention blocks (ablation) | TransMVSNet FMT · DTU evaluation set | Overall (mm): 0.305 -> 0.308 | Hurt |
Closest papers by meaning
- A Global Depth-Range-Free Multi-View Stereo Transformer Network with Pose Embedding (NeurIPS 2024 · similarity 0.86)
- GoMVS: Geometrically Consistent Cost Aggregation for Multi-View Stereo (CVPR 2024 · similarity 0.85)
- MVSFormer++: Revealing the Devil in Transformer's Details for Multi-View Stereo (ICLR 2024 · similarity 0.84)
- Stereo Anywhere: Robust Zero-Shot Deep Stereo Matching Even Where Either Stereo or Mono Fail (CVPR 2025 · similarity 0.82)
- RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · similarity 0.82)
- MVSMamba: Multi-View Stereo with State Space Model (NeurIPS 2025 · similarity 0.82)
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.