search

microsoft / ProDA

Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation (CVPR 2021)
https://arxiv.org/abs/2101.10979
MIT License
286 stars 44 forks source link

Can not reproduce in Gta5 -cityscape task. #11

Closed ygjwd12345 closed 3 years ago

ygjwd12345 commented 3 years ago

We rerun your code following the default setting. The result is quite lower than reported(57.5%). our result is as follows: Overall Acc: 0.8863829080331662 Mean Acc : 0.6676762265179861 FreqW Acc : 0.8026627310122789 Mean IoU : 0.5486842157620068

super233 commented 3 years ago

Hi, I'm also interested about the reproducibility, would you please report the MIoU of each stage described in paper, and provide the corresponding logs?

panzhang0212 commented 3 years ago

please refer to #1 for the log

ygjwd12345 commented 3 years ago

Hi, I'm also interested about the reproducibility, would you please report the MIoU of each stage described in paper, and provide the corresponding logs?

Our stage3 log file is as follow run_2021_05_08_23_22_15.log

panzhang0212 commented 3 years ago

It seems that your bs is 2, while our experiments use 4. If you try bs 2, you need to change learning rate to match your bs.

ygjwd12345 commented 3 years ago

I never change the setting.

panzhang0212 commented 3 years ago

if your bs is 4, the log should be image while your log is : image that is to say, the bs is not 4, thus the total iterations change.

ygjwd12345 commented 3 years ago

if your bs is 4, the log should be image while your log is : image that is to say, the bs is not 4, thus the total iterations change.

thanks