mlberkeley / scae-pytorch Goto Github PK

hi,friend,i run your codes and all dependences make OK.the easydict download as your requirements say.but fail run .could you help me solve this problem ,my friend.

Thx for your great work.
I tried the mnist experiment with the default settings. everything went well and the rec_image was great until 26 epoch.

this is the traceback message.
Epoch 26: 46%|████▌ | 216/469 [00:20<00:24, 10.32it/s, loss=inf, v_num=sbr8, rec_log_likelihood=-1.48e+3, temp_l1=0.251, rec_mse=0.00765]
File "/home/wangjilong/scae-pytorch/scae/models/pcae.py", line 100, in training_step
raise ValueError('loss is nan')

Is it a unfixed bug? I see there's a todo comment.

Hello, How can I get the reconstructed image？Thank you