This is the tensorflow2 implementation of dualpath embedding model for image-text matching for Flickr30k dataset. You will need the GoogleNews word2vec model (Dowload) and Flickr30k Dataset (Download) to run.
Change parameter in config_dualpath.py file, then run run_train_dualpath.py file.
- tensorflow 2.0.0 (or tensorflow-gpu 2.0.0)
- tqdm
- nltk
- sklearn
- cv2
- PIL
- gensim
To download file from Drive, you can use gdown (install by pip install gdown
)
Syntax: gdown https://drive.google.com/uc?id=[FILEID]
- word2vec id: 1lX6iq6_TfngYZKUhJoppEWhqzkS30Dhc
- flickr30k id: 12KSjtMLt5gL23aNlqZLigf6jYkjo3Svt
You can find the original paper here