search

mit-han-lab / gan-compression

[CVPR 2020] GAN Compression: Efficient Architectures for Interactive Conditional GANs
Other
1.1k stars 148 forks source link

About Select the Best Model (evolution_search.py) #103

Closed Zhaoyu321 closed 2 years ago

Zhaoyu321 commented 2 years ago

Sorry for the interruption！I successfully completed the training of the supernet on my own dataset. When I proceeded to the next stage (e.g run evolution_search.py), I find that when I changed the input size to 272x880，the code cannot continue to run, like this:

Start Evolution... Sample : 0%| | 0/100 [00:00<?, ?it/s] however, when I changed the input size to 256x256 (same size as paper)，the code can works fine. Note that, I use the size of 272X880 successfully completed the first stage training and supernet training, just run evolution_search.py that the problem arose. I tried changing some parameter settings and codes, but it didn't work. May I ask if it is only limited to a fixed input size of 256 or there are other reasons？ Sincere Regards！

lmxyy commented 2 years ago

Hi! I suppose this is just because your model needs more MACs when the resolution is higher. In this case, the previous budget constraint may be infeasible for this resolution. What you could do is try to increase the budget to see if it could work or not with --budget $MACS.

Zhaoyu321 commented 2 years ago

Hi! I suppose this is just because your model needs more MACs when the resolution is higher. In this case, the previous budget constraint may be infeasible for this resolution. What you could do is try to increase the budget to see if it could work or not with --budget $MACS.

Thank you very much! It is just the budget that causes the problem. I increase the budget and the code can continue run.