NiFangBaAGe / CoordFill

[AAAI 2023] CoordFill: Efficient High-Resolution Image Inpainting via Parameterized Coordinate Querying
BSD 3-Clause "New" or "Revised" License
83 stars 7 forks source link

About train_dataset and val_dataset. #2

Closed BigConsin closed 1 year ago

BigConsin commented 1 year ago

Hi,thanks for your great job here,and share your amazing codes here. I am little confused about your dataset in your paper,said "We use the irregular mask from the previous method (Liu et al. 2018) for mask generation with holes up to 25% following HiFill (Yi et al. 2020)." But i can not find the setting in these papers or their codes.Would you please share your train and test dataset details? That would really helps us a lot. Looking forward to your reply!

NiFangBaAGe commented 1 year ago

Thanks for your interesting in our work. The irregular mask (Liu et al. 2018) can be found at https://nv-adlr.github.io/publication/partialconv-inpainting. And the setting of mask ratio for high-resolution image inpainting can be found in Section 4.3 (Yi et al. 2020). As for the code, you can generate custom dataset via creating a mask-generator function using OpenCV to generate the desired mask.

BigConsin commented 1 year ago

totally understand,thanks lot!