demo/parse_captions.py

import os
import csv
import numpy as np

img_folder = '/playpen-ssd/adyasha/projects/StoryGAN/pororo_png/'
def get_captions_by_split():

    video_len = 4
    descriptions_original = np.load(os.path.join(img_folder, 'descriptions.npy'), allow_pickle=True,
                                         encoding='latin1').item()
    followings = np.load(os.path.join(img_folder, 'following_cache4.npy'))

    train_ids, val_ids, test_ids = np.load(os.path.join(img_folder, 'train_seen_unseen_ids.npy'), allow_pickle=True)
    filenames = ['descriptions_train.csv', 'descriptions_val.csv', 'descriptions_test.csv']

    for ids, filename in zip([train_ids, val_ids, test_ids], filenames):
        im_ids = []
        for src_img_id in ids:
            tgt_img_paths = [str(followings[src_img_id][i])[2:-1] for i in range(video_len)]
            tgt_img_ids = [str(tgt_img_path).replace(img_folder, '').replace('.png', '') for tgt_img_path in
                           tgt_img_paths]
            im_ids.extend(tgt_img_ids)
            # captions = [descriptions_original[tgt_img_id] for tgt_img_id in tgt_img_ids]

        im_ids = list(set(im_ids))
        im_ids.sort()

        # captions = [descriptions_original[i] for i in im_ids]
        with open(os.path.join(img_folder, filename), 'w') as csvfile:
            # creating a csv writer object
            csvwriter = csv.writer(csvfile)
            for i in im_ids:
                csvwriter.writerow([i, descriptions_original[i][0]])

get_captions_by_split()