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

model test issue #40

Closed pbelfort closed 2 years ago

pbelfort commented 3 years ago

Hi... I had tried to run your code, but i had this kind of issue! When I run the code, I have no problem and it's created the checkpoint path with this argument (python run.py), but when I put the argument to test (python run.py --test), I have no problem too, but its not created the result path with the regenerated images. Do you now whats wrong with my attempt?

issue

xiankgx commented 3 years ago

For the second problem, your masks is 9 channels instead of the required 3. If your masks is stored in 3 channels, then modify code to not stack 3 copies of the mask to produce 9 channel mask.

The third problem, the code fails to find your training files. Check the code to see if your files location is as expected.

pbelfort commented 3 years ago

Thx for reply... About the second issue, where in the code i have to change?

xiankgx commented 3 years ago

For the 9 vs 3 channel mask problem, just disable this line, or make your mask single channel by taking just the first channel.

https://github.com/jingyuanli001/RFR-Inpainting/blob/master/model.py#L87

m-chaves commented 3 years ago

Hi, I'm using a 3 channel png image and mask, but I get the following error "expected input[1, 4, 256, 256] to have 3 channels, but got 4 channels instead".

First, I got this other error: "expected input[1, 12, 256, 256] to have 3 channels, but got 12 channels instead", so I disabled the line you suggested (https://github.com/jingyuanli001/RFR-Inpainting/blob/master/model.py#L87). Nevertheless, it looks like it's adding an extra channel at some point. I'm using the Paris Street View pre-trained model.

xiankgx commented 3 years ago

Your masks is 4 channels. Make it one channel.

m-chaves commented 3 years ago

Could you provide me an example of the type of mask that should be used? The mask I'm using is available in: https://github.com/JiahuiYu/generative_inpainting/blob/master/examples/places2/case1_mask.png. The corresponding image is: https://github.com/JiahuiYu/generative_inpainting/blob/master/examples/places2/case1_raw.png.

I tried making the mask of 1 channel, but then I get other problems later, such as "RuntimeError: The size of tensor a (4) must match the size of tensor b (3) at non-singleton dimension 1".

01SHENyankun commented 2 years ago

你好，我遇到了和你一样的问题，能帮忙指导一下嘛