Zuo-Liang Zhu

E2FGVI的预训练模型加载晚于SpyNet，故虽然log中会提示加载原始SpyNet预训练模型，但是会被E2FGVI中训练后的参数覆盖掉，不影响后续测试。

This is a legacy from MMEditing. The code is actually correct, but indeed hard to read. XD

Could you show me a clip of the video result? Did you test on your customized data or our provided data? If you use your customized data, you can try...

You can enlarge the `anchor_memory` to fit the GPU with low vram capacity.

Please give more details about the error. Did you get the error from Colab or from Huggingface?