Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2501.15138 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866929687694934016 |
|---|---|
| author | li, zhizhen zhuo, tianyi Cao, Yifei Yu, Jizhe Liu, Yu |
| author_facet | li, zhizhen zhuo, tianyi Cao, Yifei Yu, Jizhe Liu, Yu |
| contents | Video stabilization often struggles with distortion and excessive cropping. This paper proposes a novel end-to-end framework, named TranStable, to address these challenges, comprising a genera tor and a discriminator. We establish TransformerUNet (TUNet) as the generator to utilize the Hierarchical Adaptive Fusion Module (HAFM), integrating Transformer and CNN to leverage both global and local features across multiple visual cues. By modeling frame-wise relationships, it generates robust pixel-level warping maps for stable geometric transformations. Furthermore, we design the Stability Discriminator Module (SDM), which provides pixel-wise supervision for authenticity and consistency in training period, ensuring more complete field-of-view while minimizing jitter artifacts and enhancing visual fidelity. Extensive experiments on NUS, DeepStab, and Selfie benchmarks demonstrate state-of-the-art performance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2501_15138 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | TranStable: Towards Robust Pixel-level Online Video Stabilization by Jointing Transformer and CNN li, zhizhen zhuo, tianyi Cao, Yifei Yu, Jizhe Liu, Yu Computer Vision and Pattern Recognition Video stabilization often struggles with distortion and excessive cropping. This paper proposes a novel end-to-end framework, named TranStable, to address these challenges, comprising a genera tor and a discriminator. We establish TransformerUNet (TUNet) as the generator to utilize the Hierarchical Adaptive Fusion Module (HAFM), integrating Transformer and CNN to leverage both global and local features across multiple visual cues. By modeling frame-wise relationships, it generates robust pixel-level warping maps for stable geometric transformations. Furthermore, we design the Stability Discriminator Module (SDM), which provides pixel-wise supervision for authenticity and consistency in training period, ensuring more complete field-of-view while minimizing jitter artifacts and enhancing visual fidelity. Extensive experiments on NUS, DeepStab, and Selfie benchmarks demonstrate state-of-the-art performance. |
| title | TranStable: Towards Robust Pixel-level Online Video Stabilization by Jointing Transformer and CNN |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2501.15138 |