RelTextGAN
Towards editing images via natural language. Trained using image captioning data.
Dependencies
$ pip install ftfy regex tqdm
$ pip install -f https://download.pytorch.org/whl/cu110/torch_stable.html torch==1.7.1+cu110 torchvision==0.8.2
$ pip install git+https://github.com/openai/CLIP.git
Data
Please download Oxford-102 and MS-COCO datasets and place them into a directory named data/
Training
To train COCO model and Oxford model respectively
python train.py [--image_size 256] [--batch_size 8] [--enc_type clip_sentence_encoding|bert_sentence_encoding]
python train_oxford.py [--image_size 256] [--batch_size 8] [--enc_type clip_sentence_encoding|bert_sentence_encoding]
Testing
python test.py --model [parent location of gen.pt model]
python test_oxford.py --model [parent location of gen.pt model]