README.md

Video Augmented Texts Data

VATEX

Each video contains 10 captions. In vatex.zip, there are:

• test/: a folder containing all available videos
• vatex_public_test_english_v1.1.json: JSON file containing all captions

Example data loading:

import os
import json

path = 'vatex_public_test_english_v1.1.json'
d = json.load(open(path, 'r'))

captions = {v['videoID']: v['enCap'] for v in d}

for vname in captions:
    video_path = os.path.join('test', vname+'.mp4')  # path to the video
    captions = captions[vname]  # a list of 10 str

MSR-VTT

Each video contains 1 caption. There are two files for MSR-VTT:

• MSRVTT.zip: contains all videos
• MSRVTT_JSFUSION_test.csv: contains all captions

Example data loading:

import os
import pandas as pd

path = 'MSRVTT_JSFUSION_test.csv'
df = pd.read_csv(path)

vid_id_list = df['video_id'].tolist()
caption_list = df['sentence'].tolist()

for vid_id, caption in zip(vid_id_list, caption_list):
    video_path = os.path.join('MSRVTT', 'videos', 'all', vid_id+'.mp4')
    captions = [caption]  # a list of 1 str