Haomiao Ni

I have discussed with [isht7](https://github.com/isht7), and I find that the train.py in his code sets model to be evaluate mode firstly. But I find that your code sets the model...

OK, Thanks for your reply~

@siinem You can just set model.eval() to freeze the statistic. Please check the similar code from [isht7](https://github.com/isht7/pytorch-deeplab-resnet/blob/0c00ca1656dac8263b3d5db51910ee4fa0b10b49/train.py#L200).

Hi, Kongming, Thanks for your answer! But what about the meaning of value 4? @GriffinLiang

I found that this GitHub [repo](https://github.com/wyhuai/DDNM) provided a copy of `celeba_hq.ckpt`. Hope it can help some of you.

I found that this GitHub [repo](https://github.com/wyhuai/DDNM) provided a copy of `celeba_hq.ckpt`, the model checkpoint trained on the CelebA dataset. Hope it can help some of you.

Hello, I also have the same question. As the author mentioned, he referred to the code from [weakalign](https://github.com/ignacio-rocco/weakalign), I find that [here](https://github.com/ignacio-rocco/weakalign/blob/52505eaeed52dc82cf755919b119ef99fe8b1ed1/model/loss.py#L119), the offset is also set to be 227/210....

Hi, @haofanwang, sorry for the late reply! I have created a PR for adding our CVPR 23&24 paper about image-to-video generation. Please feel free to let me know if you...

Hi, @alasokolova, thanks a lot for your interest in our work. Depending on the complexity of the new subject movement, It is hard to answer how much data is needed...

TI2V-Zero: Zero-Shot Image Conditioning for Text-to-Video Diffusion Models 论文/Paper: https://arxiv.org/pdf/2404.16306 代码/Code: https://github.com/merlresearch/TI2V-Zero