invalid value when training

JACKYLUO1991 / FaceParsing

EHANet: An effective hierarchical aggregation network for face parsing

MIT License

169 stars 31 forks source link

invalid value when training #17

Closed Chic-J closed 3 years ago

Chic-J commented 3 years ago

$M3K$@1EEVOAV8`{JA_F1R4T$

i wonder what causes this problem? besides，i wonder why the loss is so small? i train the model and the loss became 4.3e-0.5 using the FaceParseNet18 or EHANet18?

Chic-J commented 3 years ago

solved.but a new error comes up. ![Uploading image.png…]()

JACKYLUO1991 commented 3 years ago

@Chic-J Sorry, your picture doesn't appear on the page.

Chic-J commented 3 years ago

@Chic-J Sorry, your picture doesn't appear on the page.

All done. but I wonder where do you put the output images of the testing?and how can I test on my own photo?should i take one and make labels of it ?

Chic-J commented 3 years ago

and the epoch is 10.but the result is 10 .pth file...this means the 9th file is the final result?

JACKYLUO1991 commented 3 years ago

@Chic-J For testing, please refer to the script in the deployment folder.

Chic-J commented 3 years ago

@Chic-J For testing, please refer to the script in the deployment folder.

OK will try it

Chic-J commented 3 years ago

I found a warning.When i choose EHANet18 mode in detection.py and run it, a fault occurs.but good with FaceParseNet18.

Chic-J commented 3 years ago

solved.