search

jingyuanli001 / RFR-Inpainting

The source code for CVPR 2020 accepted paper "Recurrent Feature Reasoning for Image Inpainting"
MIT License
355 stars 76 forks source link

bad test result #42

Open tengshaofeng opened 3 years ago

tengshaofeng commented 3 years ago

dear @jingyuanli001 , I really appreciate your great job. The result in your paper is amazing. Then I test my own image with your released pretrained model 'RFR_Net_pretrained/checkpoint_celeba.pth'. But my result is bad. Can you help me? the input are: 21 21

if I use my own mask, the result: img_1

masked_img_1 or: img_1 masked_img_1

if use the random mask(set mask_mode=1), the result: img_1 masked_img_1

dose it depends on the masks inputed?

jingyuanli001 commented 3 years ago

using a mask with 1 indicating the hole region and 0 indicating the background should be adequate. But I am not sure if the result is going to be good when the image is not from the same distribution as CelebA

tengshaofeng commented 3 years ago

using a mask with 1 indicating the hole region and 0 indicating the background should be adequate. But I am not sure if the result is going to be good when the image is not from the same distribution as CelebA

thanks for your reply. My input mask fill the hole with 0, and fill other region with 255, and set mask_reverse = False. if I set mask_reverse = True, then it means mask with 1 indicating the hole region and 0 indicating the background , but the result is worse, the output is not expected.

nachifur commented 3 years ago

@tengshaofeng I have the same problem as you, have you solved it?

Nuha1412 commented 2 years ago

Do you solve the problem ?? I need the answer please !