Large-scale Diverse Video (LDV) Dataset

Download the latest version: LDV 3.0 (365 videos) [Raw videos] [Info]

The LDV 1.0, LDV 2.0 and LDV 3.0 datasets are established for the Video Enhancement Challenges in NTIRE Workshop 2021, NTIRE Workshop 2022 and AIM Workshop 2022.

If the datasets and the benchmarks are useful for your research, please cite:

@inproceedings{yang2021dataset,
  title={{NTIRE 2021} Challenge on Quality Enhancement of Compressed Video: Dataset and Study},
  author={Ren Yang and Radu Timofte}, 
  booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops}, 
  year={2021}
}

@inproceedings{yang2022ntire,
  title={{NTIRE 2022} Challenge on Super-Resolution and Quality Enhancement of Compressed Video: Dataset, Methods and Results},
  author={Ren Yang and Radu Timofte and others}, 
  booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops}, 
  year={2022}
}

@inproceedings{yang2022aim,
  title={{AIM 2022} Challenge on Super-Resolution of Compressed Image and Video: Dataset, Methods and Results},
  author={Ren Yang and Radu Timofte and others}, 
  booktitle={European Conference on Computer Vision Workshops}, 
  year={2022}
}

Should you have any questions, please feel free to contact:

Ren Yang @ ETH Zurich, Switzerland

Email: [email protected]

LDV 1.0 (240 videos)

The proposed LDV 1.0 in NTIRE 2021 contains 240 videos with diverse categories of content, different kinds of motion and various frame-rates. The dataset may be further extended in the future. The details of the proposed LDV 1.0 dataset are discribed in the dataset report:

Ren Yang and Radu Timofte, "NTIRE 2021 Challenge on Quality Enhancement of Compressed Video: Dataset and Study", in IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2021. [Paper]

Download LDV 1.0

• Training set (200 videos) [Raw] [Compressed (fixed QP)] [Compressed (fixed bit-rate)] [Info]

• Validation set (20 videos) [Raw] [Compressed (fixed QP)] [Compressed (fixed bit-rate)] [Info]

• Test set for Tracks 1 and 2 (10 videos) [Raw] [Compressed (fixed QP)] [Compressed (fixed bit-rate)] [Info]

• Test set for Track 3 (10 videos) [Raw] [Compressed (fixed QP)] [Compressed (fixed bit-rate)] [Info]

The NTIRE 2021 challenge compresses videos in the YUV domain and evaluates results in the RGB domain. The following commands can be used to convert the videos to the YUV and RGB domains, respectively.

ffmpeg -i xxx.mkv -pix_fmt yuv420p xxx.yuv
ffmpeg -i xxx.mkv ./xxx/%3d.png

LDV 2.0 (LDV 1.0 + 95 videos = 335 videos)

The proposed LDV 2.0 in NTIRE 2022 contains 335 videos with diverse categories of content, different kinds of motion and various frame-rates. The details of the proposed LDV 2.0 dataset are discribed in the dataset report:

Ren Yang, Radu Timofte, and others, "NTIRE 2022 Challenge on Super-Resolution and Quality Enhancement of Compressed Video: Dataset, Methods and Results", in IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2022. [Paper]

Download LDV 2.0

• The whole dataset (335 videos) [Raw videos] [Info]

• Track 1

  • Training set (LDV 1.0, 240 videos): [Raw] [Compressed (Fixed QP)] [Info]
  • Validation set (15 videos): [Raw] [Compressed (Fixed QP)] [Info]
  • Test set (15 videos): [Raw] [Compressed (Fixed QP)] [Info]

• Track 2 (The sizes of raw videos are cropped to the multiples of 16)

  • Training set (LDV 1.0, 240 videos): [Raw] [Compressed (Fixed QP)] [Info]
  • Validation set (15 videos): [Raw] [Compressed (Fixed QP)] [Info]
  • Test set (15 videos): [Raw] [Compressed (Fixed QP)] [Info]

• Track 3 (The sizes of raw videos are cropped to the multiples of 64)

  • Training set (LDV 1.0, 240 videos): [Raw] [Compressed (Fixed QP)] [Info]
  • Validation set (15 videos): [Raw] [Compressed (Fixed QP)] [Info]
  • Test set (15 videos): [Raw] [Compressed (Fixed QP)] [Info]

The NTIRE 2022 challenge compresses videos in the YUV domain and evaluates results in the RGB domain. The following commands can be used to convert the videos to the YUV and RGB domains, respectively.

ffmpeg -i xxx.mkv -pix_fmt yuv420p xxx.yuv
ffmpeg -i xxx.mkv ./xxx/%3d.png

LDV 3.0 (LDV 2.0 + 30 videos = 365 videos)

The proposed LDV 3.0 in AIM 2022 contains 365 videos with diverse categories of content, different kinds of motion and various frame-rates. The details of the proposed LDV 3.0 dataset are discribed in the dataset report:

Ren Yang, Radu Timofte, and others, "AIM 2022 Challenge on Super-Resolution of Compressed Image and Video: Dataset, Methods and Results", in European Conference on Computer Vision Workshops, 2022. [Paper]

Download LDV 3.0

• The whole dataset (365 videos) [Raw videos] [Info]

• Track 2 (The sizes of raw videos are cropped to the multiples of 64)

  • Validation set (15 videos): [Raw] [Compressed (Fixed QP)] [Info]
  • Test set (15 videos): [Raw] [Compressed (Fixed QP)] [Info]

The AIM 2022 challenge compresses videos in the YUV domain and evaluates results in the RGB domain. The following commands can be used to convert the videos to the YUV and RGB domains, respectively.

ffmpeg -i xxx.mkv -pix_fmt yuv420p xxx.yuv
ffmpeg -i xxx.mkv ./xxx/%3d.png