Research map / Label-efficient and robust learning / Knowledge distillation
Optical Flow Distillation: Towards Efficient and Stable Video Style Transfer
Xinghao Chen, Yiman Zhang, Yunhe Wang, Han Shu, Chunjing Xu, Chang Xu
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 residual distillation loss (ablation) | Student from scratch · MPI Sintel 5 scenes | e_stab sum: 0.4019 -> 0.3424 | Improved |
| add low-rank distillation loss from input videos (ablation) | Student from scratch · MPI Sintel 5 scenes | e_stab sum: 0.4019 -> 0.3855 | Improved |
| add low-rank loss to residual distillation loss (ablation) | Student + L_res · MPI Sintel 5 scenes | e_stab sum: 0.3424 -> 0.3050 | Improved |
| distill low rank from input videos instead of teacher outputs (ablation) | Student + L_res · MPI Sintel 5 scenes | e_stab sum: 0.3415 -> 0.3050 | Improved |
| add temporal loss to student baseline | Student from scratch · MPI Sintel 5 scenes | e_stab sum: 0.4019 -> 0.3787 | Improved |
| add residual distillation loss on top of temporal loss (ablation) | Student w/ L_temp · MPI Sintel 5 scenes | e_stab sum: 0.3787 -> 0.3317 | Improved |
| add low-rank loss on top of temporal loss and residual loss (ablation) | Student w/ L_temp + L_res · MPI Sintel 5 scenes | e_stab sum: 0.3317 -> 0.2938 | Improved |
| apply full optical flow distillation to student with temporal loss | Student w/ L_temp · MPI Sintel 5 scenes | e_stab sum: 0.3787 -> 0.2938 | Improved |
| remove optical flow from inference via distilled student instead of teacher | Teacher · MPI Sintel 5 scenes | FPS: 4.67 -> 183 | Improved |
| distill student instead of using teacher network | Teacher · MPI Sintel 5 scenes | e_stab sum: 0.3435 -> 0.2938 | Improved |
| increase K from 3 to 4 frames for low-rank loss (ablation) | Student (Ours) · MPI Sintel 5 scenes | e_stab sum: 0.3158 -> 0.3061 | Improved |
| increase K from 4 to 5 frames for low-rank loss (ablation) | Student (Ours) · MPI Sintel 5 scenes | e_stab sum: 0.3061 -> 0.3050 | No clear effect |
Closest papers by meaning
- Coherent Online Video Style Transfer (ICCV 2017 · similarity 0.87)
- PCA-Based Knowledge Distillation Towards Lightweight and Content-Style Balanced Photorealistic Style Transfer Models (CVPR 2022 · similarity 0.87)
- Arbitrary Video Style Transfer via Multi-Channel Correlation (AAAI 2021 · similarity 0.87)
- FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis (CVPR 2024 · similarity 0.86)
- Optical-Flow Guided Prompt Optimization for Coherent Video Generation (CVPR 2025 · similarity 0.84)
- Motion Consistency Model: Accelerating Video Diffusion with Disentangled Motion-Appearance Distillation (NeurIPS 2024 · similarity 0.84)
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.
