Please visit http://xiaotong.me/static/projects/person-search-language/dataset.html and download images of CUHK-PEDES. The alternative download link is at http://cuhk-pedes.shuanglee.me/ (You may need to contact with the author to acqurie the dataset.)

The 'reid_raw.json' is from https://github.com/ShuangLI59/Person-Search-with-Natural-Language-Description. This file can be used to split the train/val/test.

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

(matlab) make_dictionary

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

(matlab) prepare_wordcnn_feature2