ProReflow: Progressive Reflow with Decomposed Velocity
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916650661445632 |
|---|---|
| author | Ke, Lei Xu, Haohang Ning, Xuefei Li, Yu Li, Jiajun Li, Haoling Lin, Yuxuan Jiang, Dongsheng Yang, Yujiu Zhang, Linfeng |
| author_facet | Ke, Lei Xu, Haohang Ning, Xuefei Li, Yu Li, Jiajun Li, Haoling Lin, Yuxuan Jiang, Dongsheng Yang, Yujiu Zhang, Linfeng |
| contents | Diffusion models have achieved significant progress in both image and video generation while still suffering from huge computation costs. As an effective solution, flow matching aims to reflow the diffusion process of diffusion models into a straight line for a few-step and even one-step generation. However, in this paper, we suggest that the original training pipeline of flow matching is not optimal and introduce two techniques to improve it. Firstly, we introduce progressive reflow, which progressively reflows the diffusion models in local timesteps until the whole diffusion progresses, reducing the difficulty of flow matching. Second, we introduce aligned v-prediction, which highlights the importance of direction matching in flow matching over magnitude matching. Experimental results on SDv1.5 and SDXL demonstrate the effectiveness of our method, for example, conducting on SDv1.5 achieves an FID of 10.70 on MSCOCO2014 validation set with only 4 sampling steps, close to our teacher model (32 DDIM steps, FID = 10.05). |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_04824 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | ProReflow: Progressive Reflow with Decomposed Velocity Ke, Lei Xu, Haohang Ning, Xuefei Li, Yu Li, Jiajun Li, Haoling Lin, Yuxuan Jiang, Dongsheng Yang, Yujiu Zhang, Linfeng Graphics Artificial Intelligence Computer Vision and Pattern Recognition Diffusion models have achieved significant progress in both image and video generation while still suffering from huge computation costs. As an effective solution, flow matching aims to reflow the diffusion process of diffusion models into a straight line for a few-step and even one-step generation. However, in this paper, we suggest that the original training pipeline of flow matching is not optimal and introduce two techniques to improve it. Firstly, we introduce progressive reflow, which progressively reflows the diffusion models in local timesteps until the whole diffusion progresses, reducing the difficulty of flow matching. Second, we introduce aligned v-prediction, which highlights the importance of direction matching in flow matching over magnitude matching. Experimental results on SDv1.5 and SDXL demonstrate the effectiveness of our method, for example, conducting on SDv1.5 achieves an FID of 10.70 on MSCOCO2014 validation set with only 4 sampling steps, close to our teacher model (32 DDIM steps, FID = 10.05). |
| title | ProReflow: Progressive Reflow with Decomposed Velocity |
| topic | Graphics Artificial Intelligence Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2503.04824 |