Dimension ordering in 'models/context_cluster.py'

[ujaejoon]

Hi,

I've been examining the models/context_cluster.py file in your GitHub repository and observed that the ordering of the height (H) and width (W) dimensions in some tensor operations, such as adaptive pooling (nn.AdaptiveAvgPool2d) and data rearrangement (rearrange), seems unconventional. Although the input images are square, which might not cause practical issues, I am curious to know whether this ordering is intentional for algorithmic or optimization reasons, or if it was an error in implementation.

Thank you.