search

lihaod / Deep_inpainting_localization

Implementation of “Localization of Deep Inpainting Using High-Pass Fully Convolutional Network”
MIT License
27 stars 2 forks source link

Question on the dataset structure #4

Closed sejmoonwei closed 2 years ago

sejmoonwei commented 3 years ago

I set 'data_dir' = ./train The folder train is organized as : – train   |– JPG     |–1.jpg     |–2.jpg     ……   |– MASK     |–1.png     |–2.png     …… as python raises error, I guess there's something wrong with the traindata. Could you please specify the correct structure for the train folder? Do appreciate your great work

xueyaosmile commented 2 years ago

I set 'data_dir' = ./train The folder train is organized as : – train   |– JPG     |–1.jpg     |–2.jpg     ……   |– MASK     |–1.png     |–2.png     …… as python raises error, I guess there's something wrong with the traindata. Could you please specify the correct structure for the train folder? Do appreciate your great work

hi, I encountered the same problem as you, how did you solve it?

lihaod commented 2 years ago

To simplify the organization of data, I put the images and the corresponding masks into similar paths. The only difference is that the image path contains ‘jpg’ while the mask path contains ‘msk’. For example: ./ ├─train/ │ ├─jpg96/ │ │ ├─a/ (inpainted images produced by method a） │ │ │ ├─0001.jpg │ │ │ └─0002.jpg │ │ └─b/ (inpainted images produced by method b） │ ├─msk96/ │ │ ├─a/ (corresponding groundtruth masks for jpg96/a/） │ │ │ ├─0001.png │ │ │ └─0002.png │ │ └─b/ │ ├─jpg75/ │ │ ├─a/ │ │ └─b/ │ └─msk75/ │ ├─a/ │ └─b/ └─test/ ├─jpg96/ │ ├─a/ │ └─b/ ├─msk96/ │ ├─a/ │ └─b/ ├─jpg75/ │ ├─a/ │ └─b/ └─msk75/ ├─a/ └─b/

You can modify Line 188-197 in hp_fcn.py for adapting to customized data.