Dataset of MSU HDR Video Reconstruction Benchmark: HLG

Dataset preview

In this section you can observe the frames from the videos included in the dataset. All images were processed by tonemapper.

1 / 20
2 / 20
3 / 20
4 / 20
5 / 20
6 / 20
7 / 20
8 / 20
9 / 20
10 / 20
11 / 20
12 / 20
13 / 20
14 / 20
15 / 20
16 / 20
17 / 20
18 / 20
19 / 20
20 / 20

Red Square
Rolex
Candles
Red Square 2
Playing Cards
Flowers
Eternal flame
GUM
Puddle
Road sign
Dog
Red Bull
Thaw
Red Bull 2
Football
Luminaire
Firework
Cola
Cola 2
Bonsai

Camera

We used a Fujifilm XT-4 camera for filming.

SDR* – tonemapped HDR video

Selection

We shot 100 different scenes to create the dataset. The video selection process for the final version of the dataset can be divided into several steps:

  • Manual selection. At this step, we have selected visually different scenes, so that the objects, the amount of movement and the type of lighting are different.
  • Feature calculation. We consider the following features:


  • After calculating the features, we selected videos to get an uniform distribution.
Below you can see the distribution of features for all captured videos

Preprocessing

After shooting, all videos are transcoded with Apple ProRes 422 HQ codec. We do not use raw files as the ProRes losses are insignificant for our task. We can easily work with such files in Davinci Resolve.
Unless otherwise specified, we always use Apple ProRes 422 HQ in this work, including for storing method results.

Tone Mapping

As in other works on the topic of Inverse Tone Mapping, we need to get SDR versions of HDR video. We have tested traditional tonemappers such as Reinhard, Mantiuk, Drago, and others. We have also collected statistics on articles in the field, and it turned out that the most popular method is Reinhard. However, its results were of insufficient quality for us. Therefore, we gave up using traditional approaches.
We settled on the HDR Tools utility integrated into FinalCut 10.6. It has no disadvantages of Reinhard and others:

  • It is fully automatic and does not require any manual adjustments, except for the input and output gamma curve
  • It gives a significantly more natural-looking video.

We did not conduct subjective research on this topic and we have found no other studies showing that HDR Tools is the best tonmapping method. However, perhaps we will do that research in the future.

Download

SDR version of the Dataset is available here.
Note that the HDR version of Dataset have not been published and will not be published in the future.

10 May 2022
See Also
MSU Image- and video-quality metrics analysis
Description of a project in MSU Graphics and Media Laboratory
Super-Resolution Quality Metrics Benchmark
Discover 66 Super-Resolution Quality Metrics and choose the most appropriate for your videos
Video Saliency Prediction Benchmark
Explore the best video saliency prediction (VSP) algorithms
Super-Resolution for Video Compression Benchmark
Learn about the best SR methods for compressed videos and choose the best model to use with your codec
Metrics Robustness Benchmark
Check your image or video quality metric for robustness to adversarial attacks
Video Upscalers Benchmark
The most extensive comparison of video super-resolution (VSR) algorithms by subjective quality
Site structure