a problem about demo_GFC_training.m

[994088895]

I have encountered a problem when I ran the demo_GFC_training.m, the output always be the same color in a small phase, and the parsing always orange color . I dont't know the reason about it. I send you a picture about it. Thank you!

[Yijunmaverick]

@994088895

Em...looks weird. It seems that both the output and GT parsing is completely wrong. My testing looks good.

Would you first test my initial Model_G (trained with L1 only) and parsing model, separately? At least I think GT parsing should be right. Check if some experimental settings are modified, e.g. the input value range (should be between [-1,1]).

[994088895]

I only changed the training set and then cropped it to 128*128, I didn't change anything else. The input value also be between [-1,1]. I'm so confused.

[Yijunmaverick]

@994088895 I think the problem is the output of Model_G. Since the output is always white, the parsing is not right for sure.

Did you download my initial Model_G that trained with the reconstruction only?

[994088895]

I download the model follow your instruction about training , that is the Face128_Model_G_recons.caffemodel and Model_parsing.caffemodel. And the output not always white, it will change another color such as black, red and so on in a small time.the parsing is alwas orange color.

[994088895]

Can you resend me your implementation about training to my email [email protected], thank you for helping me. I can't find what's wrong with me.

[Yijunmaverick]

I just download my code and re-run it on my machine. The result in the first few iterations is shown below. I'm not sure how to comment on your issue. How about trying my test code first to see whether the result is right?

[994088895]

The test code is right to run on my machine with your final face completion model.

[994088895]

it's so wired, the dataset is just cropped to 128*128, no other preprocess on it?

[Yijunmaverick]

Okay, what the result will be if you use my final model to start the training?

The pre-processing step is already included in the code.

I guess you can try to debug by setting breakpoints in Matlab, right?

[994088895]

I just use your final model to test, not training. I will try to debug it, thank you.

[DecenWill]

I fixed it. :)

[Yuanyuan0218]

@DecenWill @994088895 it looks you guys run this successfully. but i have a lot of problems running this, so can you provide a email or we chat or other things that i can contact you for further information?thank you