Dataset Card for VideoEval

VidTAB

Action Recognition in Dark

You could download all videos from ARID at https://opendatalab.com/OpenDataLab/Action_Recognition_in_the_Dark.

You just need to use the mp4 video in the video folder and then use the annotations we provided.

Action Recognition in Long Video

You could download all videos from Breakfast at https://serre-lab.clps.brown.edu/resource/breakfast-actions-dataset/.

You just need to use the mp4 video in the video folder and then use the annotations we provided.

Medical Surgery

You could download all videos from SurgicalActions160 at http://ftp.itec.aau.at/datasets/SurgicalActions160/index.html.

You just need to use the mp4 video in the video folder and then use the annotations we provided.

Animal Behavior

You could download all videos from Animal Kingdom at https://forms.office.com/pages/responsepage.aspx?id=drd2NJDpck-5UGJImDFiPVRYpnTEMixKqPJ1FxwK6VZUQkNTSkRISTNORUI2TDBWMUpZTlQ5WUlaSyQlQCN0PWcu.

You just need to use the mp4 video in the video folder and then use the annotations we provided.

Harmful Content

You could download all videos from MOB at https://drive.google.com/file/d/1Zjib-WaF5hk3wVrj5eW6ewdpMFcn45Wo/view.

Merge folders benign and malicious and then use the annotations we provided.

Fake Face

You could download all videos from FaceForensics++ at https://docs.google.com/forms/d/e/1FAIpQLSdRRR3L5zAv6tQ_CKxmK4W96tAab_pfBu2EKAgQbeDVhmXagg/viewform?pli=1.

Then

cd yourpath/FaceForensics++
mkdir videos
mv faceforensics_videos/original_sequences/youtube/c23 videos/pos
mkdir videos/neg
python get_negs_samples.py

get_negs_samples.py is

import os
import shutil

video_list = os.listdir('videos/pos')
assert len(video_list) == 1000, len(video_list) 

for i in range(0, 1000):
    for method in ["Deepfakes", "Face2Face", "FaceShifter", "FaceSwap", "NeuralTextures"]:
        shutil.copy(f"faceforensics_videos/manipulated_sequences/{method}/c23/videos/{video_list[i]}", f"videos/neg/{video_list[i][:-4]}-{method}.mp4")

And then use the annotations we provided.

Emotion Analysis

You could download all videos from CAER at https://drive.google.com/file/d/1JsdbBkulkIOqrchyDnML2GEmuwi6E_d2/view

You just need to use the mp4 video in the video folder and then use the annotations we provided.

Quality Access

You could download all videos from DOVER at https://huggingface.co/datasets/teowu/DIVIDE-MaxWell/resolve/main/videos.zip.

You just need to use the mp4 video in the video folder and then use the annotations we provided.

VidEB

FIVR-5K

• Install yt-dlp (make sure it is up-to-date)
• Run the following command to download videos:

python VidEB/annotations/FIVR-5K/download_dataset.py \
                           --video_dir VIDEO_DIR \
                           --dataset_ids VidEB/annotations/FIVR-5K/used_videos.txt \
                           --cores NUMBER_OF_CODES \
                           --resolution RESOLUTION

DVSC23

For queries,

wget -i VidEB/annotations/DVSC23/vsc_queries.txt --cut-dirs 2 -x -nH

For database,

wget -i VidEB/annotations/DVSC23/vsc_database.txt --cut-dirs 2 -x -nH

Citation

BibTeX:

@article{li2024videoeval,
  title={Videoeval: Comprehensive benchmark suite for low-cost evaluation of video foundation model},
  author={Li, Xinhao and Huang, Zhenpeng and Wang, Jing and Li, Kunchang and Wang, Limin},
  journal={arXiv preprint arXiv:2407.06491},
  year={2024}
}

Dataset Card Contact

xinhaoli00@outlook.com