In the training and validation phase, all the input are image crops(100*100)，and the result are decent. Why is the result very poor during the test phase?

aiff22 / DPED

Software and pre-trained models for automatic photo quality enhancement using Deep Convolutional Networks

1.68k stars 369 forks source link

In the training and validation phase, all the input are image crops(100*100)，and the result are decent. Why is the result very poor during the test phase? #11

Closed qicathetine closed 6 years ago

aiff22 commented 6 years ago

Hi @qicathetine

The results during the validation and test phases should be the same. Try to test the provided pre-trained models to make sure that your test code is correct, or please provide more details about your issue if you need further help.

qicathetine commented 6 years ago

Hi@aiff22, I think the idea you provided to me is correct that the results during the validation and test phases should be consistent. Therefore, I think there is a problem with my test code. So, I would like to ask you a question. How was the image tested in the network during the test phase? Whether it is the entire picture or the divided image blocks just like the way of the training phase. Finally get the final results of combining them. Thank you!

aiff22 commented 6 years ago

Hi @qicathetine

The entire image is passed to a neural network during the test phase. Please have a look at our test code provided within this repository that contains all necessary steps for producing the final visual results: https://github.com/aiff22/DPED/blob/master/test_model.py

qicathetine commented 6 years ago

Hi@aiff22 Thank your reply, I got where the problem is.