Generate_dataset_using_GAN_and_stitching

Generate a dataset of cracked phones using GANs (to generate the cracks) and stitching the cracks on normal phone images.

Requirements

Pytorch
Scipy

Training/Generating

To train the GAN on custom data run the below command
python dc_gan_pytorch.py --data_train_path='path/to/training/images' --model_save_path='path/to/save/model'

To generate the images run the below command
python generate_images_dc_gan.py --data_save_path='path/to/training/images' --model_save_path='path/to/save/model'

To stitch the generated patches on images run
python stitch_images.py --out_dir='path/to/save/images' --input_image_dir='path/to/input/images' --input_patch_dir='path/to/input/patches' --out_csv_file_path='path/to/save/labels'