This project generates labeled images datasets by different GANs. Anyone can add his own dataset and train GAN model.
Following these steps and you'll have your own generated dataset.
- Step 1: Insert your dataset images in dataset/train_images with number folder for each label. For example if you have 3 labels, label 1 will be called 1 and this folder will contain all the images from label 1.
- Step 2: Choose your relevant arguments:
- m = model (string) DCGAN or WGAN
- e = epochs (int) not required
- gi = generated iterations (int) not required
- bs = batch_size (int)
- oi = output_items (int) - The number of items that you want to get from the generator.
- nor = number_of_rows (int) - How many rows that you want for the output images.
- ims = images_size (int) - How to normalize the input images.
- cb = create_batches (string) - Set False if you don't want to create new batches.
- t = train (string) - Set False if you just want to generate new images by exist models.
Running training of DCGAN model
python main.py --model DCGAN
--bs 200
--oi 25
--nor 5
--ims 64
--epochs 30
--t True
--cb True
--e True
