search

blackprotoss / GSDM

Text Image Inpainting via Global Structure-Guided Diffusion Models (Accepted by AAAI-24)
MIT License
46 stars 4 forks source link

please explain what data is needed for training the SPM in the paper #9

Open Alice77bai opened 1 month ago

Alice77bai commented 1 month ago

Could you please explain what data is needed for training the SPM in the paper? The supplementary materials mention datasets including CI: Corrupted Images, II: Intact Images, CM: Corrupted Masks, ISM: Intact Segmentation Masks, CSM: Corrupted Segmentation Masks, IG: Intact Glyphs, and CC: Character Coordinates. However, the train_spm config only loads two types of images. Which two should be used? Should it be CI: Corrupted Images and II: Intact Images, or CI: Corrupted Images and ISM: Intact Segmentation Masks?

blackprotoss commented 1 month ago

we offer diverse information for the subsequent in-depth research. In this study, we solely utilized Corrupted Images (as input) and Intact Segmentation Masks (as output) for the training of SPM.