Low metrics

[OptimusFaber]

Hello!
Thank you for your work! I've tried to run your code and faced a lot of syntax mistakes, but after all I managed to start training and validation, but I got very low metrics, which are different from yours in your paper. I haven't changed anything fundamental in your code, so can you please check what the problems can be? I see only 3 options:

1. Wrong parameters,
2. Wrong code,
3. I messed something up while fixing the code.
   I forked you repo - it's in https://github.com/OptimusFaber/ActiveSGM
   What i did:
   1 I wrote new setup conda script (previous wasn't working),
   2 I changed dimension issues (after starting training - script was throwing errors on about dimension's sizes),
   3 I returned parts for cuda cache cleaning,
   4 I added validation during training (in your adaptation it was only in the end).

[WW11111]

Excuse me！Could you please perform a semantic evaluation at the end? My semantic ground truth and the rendered do not match. The results are shown below, with the semantic truth value on the far left.