Research map / Architectures and efficiency / RNNs and normalization
FreeFlow: A Bias-free Hierarchical Transformer for Optical Flow Estimation
Vladislav Bargatin, Alexander Yakovenko, Khaled Abud, Dmitriy Vatolin
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 shifted-window attention (Win+Global to Win+Swin+Global, mask 0.9) (ablation) | FreeFlow (small) · Spring sub-val | 1px: 0.801 -> 0.688 | Improved |
| Add Global attention at masking ratio 0.9 (ablation) | FreeFlow (small) · Spring sub-val | 1px: 0.659 -> 0.688 | No clear effect |
| Add Global attention at masking ratio 0.95 (ablation) | FreeFlow (small) · Spring sub-val | 1px: 0.658 -> 0.624 | Improved |
| Raise masking ratio from 0.9 to 0.95 (ablation) | FreeFlow (small) · Spring sub-val | 1px: 0.688 -> 0.624 | Improved |
| Raise masking ratio from 0.95 to 0.975 (ablation) | FreeFlow (small) · Spring sub-val | EPE: 0.157 -> 0.174 | Hurt |
| Use 16x16 patches instead of 8x8 (ablation) | FreeFlow-S · Spring sub-val | 1px: 0.709 -> 0.798 | Hurt |
| Replace local-global attention with vanilla ViT (ablation) | FreeFlow-S · Spring sub-val | 1px: 0.709 -> 0.698 (ViT 12 layers, ~6x slower) | Hurt |
| Add GeoViT-like iterative warping to FreeFlow (ablation) | FreeFlow-S · Spring sub-val | 1px: 0.709 -> 0.583 | Improved |
| Scale from FreeFlow-S to L (ablation) | FreeFlow · Spring test | EPE: 0.533 -> 0.278 | Improved |
| Increase FreeFlow size zero-shot (ablation) | FreeFlow · KITTI-15 (train) zero-shot | Fl-all: 10.4 (S) / 14.6 (M) / 12.9 (L) | Mixed |
Closest papers by meaning
- GMFlow: Learning Optical Flow via Global Matching (CVPR 2022 · similarity 0.88)
- CRAFT: Cross-Attentional Flow Transformer for Robust Optical Flow (CVPR 2022 · similarity 0.88)
- StreamFlow: Streamlined Multi-Frame Optical Flow Estimation for Video Sequences (NeurIPS 2024 · similarity 0.88)
- MemFlow: Optical Flow Estimation and Prediction with Memory (CVPR 2024 · similarity 0.87)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.87)
- DPFlow: Adaptive Optical Flow Estimation with a Dual-Pyramid Framework (CVPR 2025 · 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.
