Poor quality image inpainting

[asideras]

Hello everyone, congratulations on your great work.

I am attempting to use a pre-trained model to perform image inpainting on my own images. However, I am obtaining poor results for multiple images. Below, you can see the masked image, the mask, the output, and the code I executed:

!python demo/inpainting_demo.py \ configs/deepfillv2/deepfillv2_8xb2_places-256x256.py \ https://download.openmmlab.com/mmediting/inpainting/deepfillv2/deepfillv2_256x256_8x2_places_20200619-10d15793.pth \ tests/data/inpainting/Places365_val_00000012_masked.jpg \ tests/data/inpainting/Places365_val_00000012_mask.jpg \ tests/data/pred/RES.png

What did I do wrong?

Also, later on, I would like to run the model on multiple images (a subset of Places365) and use my own masks. Is this possible?
Thanks in advance.

[Z-Fran]

We have supported new demo here #1833 . You can run:

python demo/mmagic_inference_demo.py \
        --model-name deepfillv2  \  
        --img ./resources/input/inpainting/celeba_test.png \
        --mask ./resources/input/inpainting/bbox_mask.png \
        --result-out-dir ./output.jpg \
        --model-setting 1

[Z-Fran]

Closing this issue since the bug has been fixed in a merged pull request. Please feel free to reopen it if it has not been fixed.