search

jianzhangcs / panini

Panini-Net: GAN Prior based Degradation-Aware Feature Interpolation for Face Restoration, AAAI 2022 (PyTorch Code)
112 stars 13 forks source link

testing your model made bad quality images. #4

Open vemiemi opened 2 years ago

vemiemi commented 2 years ago

Thank you for giving us expressive your work.

When i tried your code on colab, output image quality is not so good.

Is there anything i have to do for getting good quality?

input and output are like below. input(SR model) a0004038_main____ output(SR model) 20a18724-abaa-40c6-8b6f-0092ccdbf563

input(MFR model) 2a1bef3d-2faa-42a0-ae8a-a7a00f9b7d6b output(MFR model) 190790d6-e4de-4963-a5d9-adfe486d9143

jianzhangcs commented 2 years ago

Thank you for your attention and questions! This work focus on how to adapt to multiple degradation, which is currently artificially designed, so our generalization performance is not good enough. The pre-trained model provided is trained based on the degradation range given by us. The poor result of your test may be because your test image is not in the degradation domain we trained. If you want to reproduce the results of the paper, you can use the test examples we provide or use our degrade function to generate the test images.

BertyWooster commented 2 years ago

@vemiemi you can try to transform the image to the original pose, that was used in the training procedure. You can get it from samples in repo. You can see the transformation and results in the files attached. image_2022-04-26_21-37-34 image_2022-04-26_21-39-44 fem res