search

yassouali / CCT

:page_facing_up: Semi-Supervised Semantic Segmentation with Cross-Consistency Training (CVPR 2020).
https://yassouali.github.io/cct_page/
MIT License
393 stars 57 forks source link

Question about the paper "Semi-Supervised Semantic Segmentation with Cross-Consistency Training" #68

Closed Hugo-cell111 closed 1 year ago

Hugo-cell111 commented 1 year ago

Hi! Could I ask how the images in row(d) from Figure 2 are drawn? The resolution of the intermediate representations seems not the same as the resolution of input images. In other word, how can I expand the resolution of the hidden representations to the same shape as the input images? Thanks!

yassouali commented 1 year ago

Hi @Hugo-cell111

You can either upsample the features then compute the distances or compute the distances then upsample them, the upsampling is only done for better visualization, you can even just apply TSNE / PCA over the features to see the clustering of the features

Hope this helps