search

terarachang / ACCV_TinyGAN

BigGAN; Knowledge Distillation; Black-Box; Fast Training; 16x compression
75 stars 9 forks source link

Hello, this error is reported step by step and cannot be solved #3

Closed KivenGood closed 3 years ago

KivenGood commented 3 years ago

Hello, this error is reported step by step and cannot be solved. In file "/pcb/TinyGAN/solver.py", line 135, 2020-12-30 12-33-29屏幕截图 2020-12-30 12-32-35屏幕截图

terarachang commented 3 years ago

Hi,

Let me explain this part of the code. First, the model will try to ask for a batch of data from ImageNet; if failed, which always happens at the very first time you launch this program as well as every time it uses up/trains on all the data in ImageNet, it would go into except to reload the real_iter.

I'm not sure why your code failed, but it is highly possible that there is something wrong with your real_loader. Please check line 25 & 134 to see if real_loader is correctly linked to the ImageNet data. If not, please check line 30 in main.py, specifically, to check what your config.real_dir is pointing to. You can see my reply to #issue2 for more details about how to organize the ImageNet data. Or, you can simply disable all the lines concerning x_real2, label2, which will lead to some drop in performance, but the program is still runnable (please see more in the ablation study in the paper).

Thank you, Ting-Yun Chang