search

sysu19351158 / FPB-FOMM

MIT License
7 stars 0 forks source link

bug #1

Closed XionghuiYe closed 4 months ago

XionghuiYe commented 4 months ago

作者你好，你发布的代码中的FramesDataset的out['driving']和out['source']都是numpy数组的形式，可是模型对输入的要求是张量类型，这地方是不是出错了，请解释下，谢谢

XionghuiYe commented 4 months ago

还有就是在该dataset中我没看到关于将视频数据转为灰度图的相关操作，是不是漏发了其它dataset的py文件？

zylye123 commented 4 months ago

Hi. Sorry for the late reply.

The dataloader construction part is dependent on the implementation of first order motion model (FOMM). You may refer to that repo.

As for the grayscale image issue. This repository is targeted at the micro-expression generation task. The SAMM dataset, which is a commonly used micro-expression dataset itself is grayscale. So we use grayscale images for demonstration here.

XionghuiYe commented 4 months ago

Hi. Sorry for the late reply.

The dataloader construction part is dependent on the implementation of first order motion model (FOMM). You may refer to that repo.

As for the grayscale image issue. This repository is targeted at the micro-expression generation task. The SAMM dataset, which is a commonly used micro-expression dataset itself is grayscale. So we use grayscale images for demonstration here.

Thank you for taking the time to reply amidst your busy schedule. I think the second issue has been resolved, but as for the first issue, after reading the FOMM documentation, I feel that it does not mention any operations related to converting numpy data into tensor types suitable for neural network training. Could you please explain why this is the case?

XionghuiYe commented 4 months ago

Hi. Sorry for the late reply.

The dataloader construction part is dependent on the implementation of first order motion model (FOMM). You may refer to that repo.

As for the grayscale image issue. This repository is targeted at the micro-expression generation task. The SAMM dataset, which is a commonly used micro-expression dataset itself is grayscale. So we use grayscale images for demonstration here.

If there are operations to convert numpy to tensor, please find them and show them to me. I would greatly appreciate it.

zylye123 commented 4 months ago

Hi.

Apparently, the transform operation is automatically done within the PyTorch.

For Chinese readers, you may also refer to this blog.

XionghuiYe commented 4 months ago

Hi.

Apparently, the transform operation is automatically done within the PyTorch.

For Chinese readers, you may also refer to this blog.

Thank you very much for your patient response! It helps me to make up for my lack of knowledge in PyTorch. Dalao，Niubee(QWQ)！