Research map / 3D vision / Depth and stereo
A Study of Finetuning Video Transformers for Multi-view Geometry Tasks
Huimin Wu, Kwang-Ting Cheng, Stephen Lin, Zhirong Wu
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 video-pretrained encoder instead of random init (ablation) | GeoViT · Sintel final (train) | EPE: 13.49 -> 1.78 | Improved |
| use spatiotemporal MAE (MAE st) instead of image MAE (ablation) | GeoViT · Sintel final (train) | EPE: 1.99 -> 1.78 | Improved |
| use MAE st instead of semantics-rich video models (MVD, UMT, InternVideo) (ablation) | GeoViT · Sintel final (train) | EPE: 1.87 -> 1.78 | Improved |
| add iterative refinement to the linear decoder (ablation) | GeoViT · Sintel final (train) | EPE: 2.00 -> 1.78 | Improved |
| increase refinement iterations from 6 to 12 (ablation) | GeoViT · Sintel final (train) | EPE: 1.78 -> 1.78 | No clear effect |
| increase encoder size from ViT-Small to ViT-Large (ablation) | GeoViT · Sintel final (train) | EPE: 2.32 -> 1.82 | Improved |
| refine by warping images instead of querying cost volume (ablation) | GeoViT · Sintel final (train) | EPE: 1.99 -> 1.78 | Improved |
| use video ViT as context encoder in RAFT (ablation) | GeoViT · Sintel final (train) | EPE: 2.71 -> 1.99 | Improved |
| use video ViT as context encoder in RAFT (ablation) | GeoViT · KITTI-15 (train) | F1-epe | No clear effect |
| apply GeoViT to depth estimation versus GMDepth | GeoViT · Scenes11 depth | Abs Rel: 0.050 -> 0.118 | Hurt |
Closest papers by meaning
- Unifying Flow, Stereo and Depth Estimation (TPAMI 2023 · similarity 0.89)
- Streaming Visual Geometry Transformer (ICLR 2026 · similarity 0.89)
- Flow3r: Factored Flow Prediction for Scalable Visual Geometry Learning (CVPR 2026 · similarity 0.89)
- Understanding Multi-View Transformers (NeurIPS 2026 · similarity 0.87)
- GTA: A Geometry-Aware Attention Mechanism for Multi-View Transformers (ICLR 2024 · similarity 0.87)
- MVSFormer++: Revealing the Devil in Transformer's Details for Multi-View Stereo (ICLR 2024 · similarity 0.86)
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.
