hzwer

Please check https://github.com/pytorch/examples/issues/467. It seems to be a OS-torch related issue.

Hi, its architecture is the same as RIFE. Did you meet any issues using this code ? https://github.com/megvii-research/ECCV2022-RIFE/blob/main/benchmark/testtime.py

Yes, the training dataset is different. I'm not sure, what is your specific configuration. In my experiments, the vimeo90K was always at 35.5-35.6dB. It is known that a learning rate...

RIFEm 可以输入 t 直接实现任意时刻插帧呀。你说的是 inference_img.py 的迭代吗，可以参照 https://github.com/megvii-research/ECCV2022-RIFE/blob/main/benchmark/HD_multi_4X.py 的推理方式。

https://github.com/megvii-research/ECCV2022-RIFE I think if you can train original RIFE, it will be not hard to finetune practical-RIFE v4.6. If you meet any problem, just raise the issue.

@wlmqslch 我修了一下 https://github.com/megvii-research/ECCV2022-RIFE/commit/58f23dc85a7f573157fa0a5447cee38742353c7e

@zzh-tech 加上ATD12k，但是没明显效果

@Q8sh2ing https://github.com/hzwer/Practical-RIFE#usage and https://github.com/megvii-research/ECCV2022-RIFE/issues/41

@Q8sh2ing https://drive.google.com/drive/folders/1lPdn7VqT-8dMG5YfXxz9zIGuBBBJKIcg?usp=sharing

@zili-zhou Hi, to support timestep, set arbitrary=True to use IFNet_m.py https://github.com/megvii-research/ECCV2022-RIFE/blob/main/model/IFNet_m.py#L63