search

hkchengrex / Mask-Propagation

[CVPR 2021] MiVOS - Mask Propagation module. Reproduced STM (and better) with training code :star2:. Semi-supervised video object segmentation evaluation.
https://hkchengrex.github.io/MiVOS/
MIT License
127 stars 22 forks source link

如何取所有内存帧中的前一帧？ #25

Closed longmalongma closed 3 years ago

longmalongma commented 3 years ago

您好，有个问题想请教下， image

这是将所有内存帧和当前帧的内存读取操作，如果说只利用当前帧的前一帧进行匹配的话，如果去内存帧的最后保留的一帧，换句话说如何取前一帧？代码中如何实现？感谢。

hkchengrex commented 3 years ago

I cannot really solve your implementation problems for you... Some hints: https://github.com/hkchengrex/Mask-Propagation/blob/b5d8e61c87e1944426c3aed43685fd0eb336353a/model/network.py#L43 tells you how the tensor is arranged. Indexing in the time dimension gives you features for different frames.

longmalongma commented 3 years ago

tells you how the tensor is arranged. Indexing in the time dimension gives you features for different frames.

这里mk形状中，B是batch？CK是特征通道数？T是帧的数目？W,H分别为像素的宽和高，这样的理解对吗？

hkchengrex commented 3 years ago

Yes.

longmalongma commented 3 years ago

Yes.

Ok, thanks!