Research map / 3D vision / Depth and stereo
CroCo v2: Improved Cross-view Completion Pre-training for Stereo Matching and Optical Flow
Philippe Weinzaepfel, Thomas Lucas, Vincent Leroy, Yohann Cabon, Vaibhav Arora, Romain Brégier, Gabriela Csurka, Leonid Antsfeld, Boris Chidlovskii, Jerome Revaud
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 |
|---|---|---|---|
| replace cosine absolute positional embedding with RoPE (ablation) | CroCo pre-training · ETH3D val stereo | bad@1.0px (%): 1.82 -> 0.60 | Improved |
| replace cosine with RoPE on MPI-Sintel flow (ablation) | CroCo pre-training · MPI-Sintel val final | EPE: 2.57 -> 2.77 | Hurt |
| add 5.3M real-world image pairs to synthetic Habitat pairs (ablation) | CroCo v2 · Middlebury val stereo | bad@1.0px (%): 25.3 -> 20.7 | Improved |
| add real data on ETH3D stereo (ablation) | CroCo v2 · ETH3D val stereo | bad@1.0px (%): 0.60 -> 0.82 | Hurt |
| scale decoder from Small to Base (ablation) | CroCo v2 · Middlebury val stereo | bad@1.0px (%): 20.7 -> 17.1 | Improved |
| scale encoder from ViT-B to ViT-L (ablation) | CroCo v2 · Middlebury val stereo | bad@1.0px (%): 17.1 -> 15.5 | Improved |
| use cross-view completion pre-training instead of random init (ablation) | CroCo-Stereo/Flow · Middlebury val stereo | bad@1.0px (%): 43.4 -> 15.5 | Improved |
| use CroCo pre-training instead of MAE ImageNet pre-training (small setup) (ablation) | CroCo · Middlebury val stereo | bad@1.0px (%): 35.8 -> 26.3 | Improved |
| use 90% masking ratio instead of 80% (ablation) | CroCo v2 pre-training · Middlebury val stereo | bad@1.0px (%): 32.5 -> 20.7 | Improved |
| tiling inference on scenes with large disparity | CroCo-Stereo · Middlebury test | avg error (px): 2.36 vs 1.89 (GMStereo) | Hurt |
| CroCo-Stereo versus best method on ETH3D bad@1.0 | CroCo-Stereo · ETH3D test | bad@1.0 noc (%): 0.98 vs 0.99 | No clear effect |
| CroCo v2 pre-trained encoder for monocular tasks versus CroCo | CroCo v2 encoder · ADE20k | mIoU: 40.6 -> 44.7 | Improved |
Later papers that compared with this method
3 later accepted papers report CroCo v2 as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- MuM: Multi-View Masked Image Modeling for 3D Vision (CVPR 2026 · better on 7 of 7 shared results)
- What VGGT Knows About Overlap: Probing Geometric Foundation Models for Co-Visibility (ECCV 2026 · better on 2 of 2 shared results)
- LoCo: Learning 3D Location-Consistent Image Features with a Memory-Efficient Ranking Loss (NeurIPS 2024 · better on 2 of 4 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CroCo: Self-Supervised Pre-training for 3D Vision Tasks by Cross-View Completion (NeurIPS 2022 · similarity 0.93)
- Cross-View Completion Models are Zero-shot Correspondence Estimators (CVPR 2025 · similarity 0.90)
- Stereo Anywhere: Robust Zero-Shot Deep Stereo Matching Even Where Either Stereo or Mono Fail (CVPR 2025 · similarity 0.88)
- Cross-view and Cross-pose Completion for 3D Human Understanding (CVPR 2024 · similarity 0.87)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.87)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 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.
