Issues with running repository on custom dataset

[vandana-sreenivasan3]

Hi, I am training the cycle Gan on a custom dataset that contains images of people wearing the jewellery and no jewellery. The images i train A and train B are of the same people with and without the jewellery. The fakes generated during the training, has people wearing the jewellery slowly. But when I use any of the training data images to do test it doesn't generate an image of the person wearing the jewellery, my dataset is relatively very small with 16 images in both directories. I am not looking for an overall very accurate model but something that can work on selective images as well. It would be really helpful if someone can please help out immediately. I am relatively new to gans. Command for test: !python test.py --dataroot /content/drive/My\ Drive/dataset/test/testA --name my_model --preprocess resize_and_crop --model test --num_test 1 --direction AtoB --no_dropout

Command for train: !python train.py --dataroot /content/drive/My\ Drive/dataset/train --name my_model --n_epochs 500 --preprocess resize_and_crop --display_id -1 --model cycle_gan --direction AtoB I used this line after train and prior test: !cp ./checkpoints/my_model/latest_net_G_A.pth ./checkpoints/my_model/latest_net_G.pth

[HassanBinHaroon]

@vandana-sreenivasan3 It seems that your dataset is too small relatively.

1. If you're not looking for a very accurate model, try replicating your training images in the val and test folders.

2. Try performing classical data augmentation techniques that are very simple and not much time-consuming.