Image-Text-Embedding/dataset/Flickr30k-prepare at master · layumi/Image-Text-Embedding · GitHub

Name		Name	Last commit message	Last commit date
parent directory ..
Flickr30k		Flickr30k
README.md		README.md
clear_txt.m		clear_txt.m
make_dictionary.m		make_dictionary.m
prepare_imdb.m		prepare_imdb.m
prepare_wordcnn_feature2.m		prepare_wordcnn_feature2.m
resize_image.m		resize_image.m
split_Flickr30k.m		split_Flickr30k.m
train_txt.m		train_txt.m

README.md

Prepare Data

1. Download Data

Please visit http://shannon.cs.illinois.edu/DenotationGraph/ and download Flickr30k.

2. Split Data set (Train/Val/Test)

(matlab) split_Flickr30k
(matlab) resize_image
(matlab) prepare_imdb  %you need to change the full path in this script

3. Clear training data and Get dicitionary

(matlab) train_txt
(matlab) make_dictionary

3. Transfer sentence to the index in dictionary

In this step, we also get rid of rare words, which are not included in GoogleNews word2vector.

(matlab) clear_txt
(matlab) prepare_wordcnn_feature2