imgs and pair folders are lmdb training data
label_en_dict.npy and label_en_graph.npz are the dictionary and graph embeddings of labels
train_texts.json and train_imgs.tsv seem to be another encoding format for the training set. Can't remember plz check the code.
two csv are summaries, just for additional information providing, not used in code
test_ex_adapt_collect.zip is the testing set, and the label_en_ex_adapt_test.txt includes the labels, respectively… See the full description on the dataset page:
https://huggingface.co/datasets/ZoeTAN/ColonCLIP.
