search

xulianuwa / MCTformer

Code for CVPR 2022 paper "Multi-Class Token Transformer for Weakly Supervised Semantic Segmentation"
152 stars 16 forks source link

MS COCO 2014 dataset #41

Open jungletada opened 11 months ago

jungletada commented 11 months ago

Nice job for Weakly Supervised Semantic Segmentation! I have a question about the segmentation on MS COCO 2014 dataset. I think that run_coco.sh doesn’t provide the steps for PSA, and also for the training part of ResNet38. Could you kindly add the rest steps for training MS COCO 2014 dataset? I think these steps include

  1. Evaluate the generated class-specific localization maps
  2. Train and evaluate the pixel semantic affinity
  3. Train and evaluate the segmentation model
J-Lam0 commented 3 months ago

Nice job for Weakly Supervised Semantic Segmentation! I have a question about the segmentation on MS COCO 2014 dataset. I think that run_coco.sh doesn’t provide the steps for PSA, and also for the training part of ResNet38. Could you kindly add the rest steps for training MS COCO 2014 dataset? I think these steps include

  1. Evaluate the generated class-specific localization maps
  2. Train and evaluate the pixel semantic affinity
  3. Train and evaluate the segmentation model

hi, do you know how to evaluate the generated class-specific localization maps for coco?