Research map / Video / Video segmentation and tracking
Two Birds, One Stone: A Unified Framework for Joint Learning of Image and Video Style Transfers
Bohai Gu, Heng Fan, Libo Zhang
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 video-image cross-attention interaction module (joint learning) (ablation) | UniST · Image style transfer | content difference D_C: 15.44 -> 12.36 | Improved |
| add video-image cross-attention interaction module (joint learning) (ablation) | UniST · Image style transfer | style difference D_S: 0.69 -> 0.46 | Improved |
| train jointly on image and video rather than image only (ablation) | UniST · Image style transfer | content difference D_C: 19.29 -> 15.44 | Improved |
| use joint training instead of sequential training (ablation) | UniST without interaction · Image style transfer | content difference D_C: 16.36 -> 15.44 | Improved |
| add more image-only training data (60K to 80K images) (ablation) | image-only UniST · Image style transfer | content difference D_C: 19.80 -> 19.98 | No clear effect |
| replace standard MSA with axial MSA (AMSA) (ablation) | UniST · efficiency | FLOPS (G): 4.29 -> 0.27 | Improved |
| replace standard MSA with axial MSA (AMSA) (ablation) | UniST · 512x512 inference | seconds per image: 0.182 -> 0.111 | Improved |
| use unimodal (image only) input at inference instead of bimodal (ablation) | UniST · Image style transfer | D_C and D_S: 12.36 -> 12.36 (D_C), 0.46 -> 0.46 (D_S) | No clear effect |
| feed output of first MSA as Q, K, V (or Q only) to the second MSA in AMSA (ablation) | AMSA · qualitative | visual quality | Hurt |
| retrain image baselines with the same training data as UniST (ablation) | CCPL, StyleFormer, IEST, Stytr2, AdaAttN · Image style transfer | D_C: UniST 12.36 vs best baseline Stytr2 12.87 | No clear effect |
Closest papers by meaning
- Learning Linear Transformations for Fast Image and Video Style Transfer (CVPR 2019 · similarity 0.89)
- Domain-Aware Universal Style Transfer (ICCV 2021 · similarity 0.88)
- UniVST: A Unified Framework for Training-Free Localized Video Style Transfer (TPAMI 2025 · similarity 0.88)
- Attention-Aware Multi-Stroke Style Transfer (CVPR 2019 · similarity 0.88)
- Semantix: An Energy-guided Sampler for Semantic Style Transfer (ICLR 2025 · similarity 0.87)
- SCSA: A Plug-and-Play Semantic Continuous-Sparse Attention for Arbitrary Semantic Style Transfer (CVPR 2025 · similarity 0.85)
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.
