SVD-download

0. About the paper

This repo is the source code for download the original videos in the paper "SVD: A Large-Scale Short Video Dataset for Near-Duplicate Video Retrieval" on ICCV-2019. The authors are: Qing-Yuan Jiang , Yi He, Gen Li, Jian Li, Lei Li and Wu-Jun Li. If you have any questions about the source code, please contact: linj#lamda.nju.edu.cn or heyi#bytedance.com.

1. Running Environment

python 3

2. Usage

python download_demo.py --help
usage: download_demo.py [-h] --dst-path DST_PATH --urls-path URLS_PATH
                        [--num-procs NUM_PROCS] [--num-retries NUM_RETRIES]
                        [--checksum-path CHECKSUM_PATH] [--verbose]

optional arguments:
  -h, --help            show this help message and exit
  --dst-path DST_PATH   destination to store videos
  --urls-path URLS_PATH
                        path to urls file
  --num-procs NUM_PROCS
                        number of process
  --num-retries NUM_RETRIES
                        number of retries
  --checksum-path CHECKSUM_PATH
                        path to checksum files
  --verbose

3. Running Demo

Pls note that the whole SVD datasets contain 562,013 videos and require about 500 GB space (488G on my device).

python download_demo.py --dst-path /path/to/dst_dir/ --urls-path /path/to/urls