search

shahaf1313 / ProCST

Official implementation of ProCST image-to-image translation for UDA-SS.
MIT License
28 stars 3 forks source link

There seems to be something wrong with your form. #2

Closed Renp1ngs closed 2 years ago

Renp1ngs commented 2 years ago

Hello, Shahaf. Thanks for your wonderful work. But I have the following problem.

  1. But it seems HRDA's mIoU on GTA->Cityscapes is 73.8, not 72.8. What do you think caused the problem of different levels of improvement under different methods? i.e. DAFormer+1.1, ProDA+0.6, HRDA+0.2
  2. Have you ever done experiments with SiT on semi-supervised segmentation tasks?

Best regards, Yanqi

shahaf1313 commented 2 years ago

Hey Yanqi, thanks for your comments.

  1. We have tested results on DAFormer and HRDA using our runs for fair comparison (hardware wise). Adittionally, we tested HRDA on 8 different seeds for robust comparison (original results were tested on 3 seeds).
  2. Unfortuitnatlly, no.
Renp1ngs commented 2 years ago

Thanks for your quick reply.

Adittionally, we tested HRDA on 8 different seeds for robust comparison (original results were tested on 3 seeds).

Solid work. Hope SiT can get into a good conference paper or journal paper.