This is an unoffical repository for reproducing model DMFN from the paper [Image Fine-grained Inpainting]. The original repository is here, but author have not commit the rest of implement code yet.
- Python3.5 (or higher)
- pytorch 1.0(or higher) with GPU
- numpy
- OpenCV
- scipy
- tensorboardX
contact us: yesure2023@gmail.com
You can specify the folder address by the option --dataset_path, and set the pretrained model path by --load_model_dir when calling test.py as the following
python test.py --dataset yesuredata/test --data_file yesuretxt/test.txt --load_model_dir checkpoints/40_net_DFBN.pth
We trained it first time 40 epochs with single GPU - GTX 1650, and also train again with advance image processing method, 50 epochs on single GTX 3060Ti, you can train it yourself for better performance or in custom dataset.
Use train.py as the following
python train.py --dataset_path data_final --data_file yesuretxt/train1.txt --batch_size 4 --lr 2e-4 --epochs 40 --load_model_dir checkpoints/40_net_DFBN.pth
You can load the pretrained model by the option --load_model_dir and edit default params in 2 python files in 'options' folder, too.