代码复现的正确率

[XZNWU]

Hello, thank you very much for sharing the code, why I use your code training only 42%

[fxle]

@xiangzhang06 Hello，When I try to train this code .The following problems are appeared.Do you know how to solve it? Thanks
/home/steven/Documents/test/DFL-CNN/main.py
DFL-CNN <==> Part1 : prepare for parameters <==> Begin
DFL-CNN <==> Part1 : prepare for parameters <==> Done
DFL-CNN <==> Part2 : Load Network <==> Begin
DFL-CNN <==> Part2 : Load Network <==> Done
DFL-CNN <==> Part3 : Load Dataset <==> Begin
Traceback (most recent call last):
File "/home/steven/Documents/test/DFL-CNN/main.py", line 180, in
main()
File "/home/steven/Documents/test/DFL-CNN/main.py", line 121, in main
dataroot = os.path.abspath(args.dataroot)
File "/usr/lib/python3.5/posixpath.py", line 357, in abspath
if not isabs(path):
File "/usr/lib/python3.5/posixpath.py", line 64, in isabs
return s.startswith(sep)
AttributeError: 'NoneType' object has no attribute 'startswith'

Process finished with exit code 1

[songdejia]

@xiangzhang06 Thank you for your reply. Specific loss weights is not '1:1:1', you can try to tune it.

[songdejia]

@fxle --dataroot /PATH/TO/YOUR_DATA

[pengshiqi]

@xiangzhang06 I got the same result as yours. Sad...
Have you met the overfitting problem? i.e. Accuracy on train set is 100% while accuracy on test set is only ~48%?

[XZNWU]

Yes, I have the same problem as you. How did you solve it ? @pengshiqi