ssivalenka / image_captioing

#Image Captioning Baseline - https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/03-advanced/image_captioning

Train: python train.py Test: python test_with_beamsearch.py --img='png/example.png' --model='BEST_checkpoint_coco_5_cap_per_img_5_min_word_freq.pth.tar' --vocab_path='data/vocab.pkl' --beam_size=5

Classes Changed: bulid vocab - Using Andrej Karpathy's caption data. As it contains entire CocoDataset in one json.

Models: encoder - changed the cnn encoder to give features instead of one single combined vector which are used for attention. decoder - added attention class

test_with_beamsearch: added this class to test an image.