Point Cloud Sequence QoE Dataset

This repository contains the results of a user study on the quality of experience of point cloud sequences. Details can be found in the corresponding publication (see references).

Data Description

The results of the study are provided by two files participants.csv and ratings.csv. We further provide details about the resource requirements with resources.csv. The following sections describe their content and the used data types.

Participants

The table participants.csv contains the following information about each participant of the user study:

Column Name	Data Type	Description
`participant`	Int	Participant ID
`gender`	String	Gender with the options: `female`, `male`, `diverse`, `omit` (prefer not to say)
`age`	String	Age, grouped by ranges (e.g. `22-23`)
`screen_type`	String	Options for used device type and screen size: `smartphone` (up to 7"), `tablet` (up to 12"), `laptop` (up to 18"), `pc` (up to 38"), `tv` (greater than 38"), `other`
`screen_avail_width`	Int	`screen.availWidth` from browser
`screen_avail_height`	Int	`screen.availHeight` from browser
`screen_width`	Int	`screen.width` from browser
`screen_height`	Int	`screen.height` from browser
`experience_games_regular_screen`	Bool	Participant has experience in "Playing video games on regular monitors / TVs"
`experience_games_smartphone`	Bool	Participant has experience in "Playing video games on handheld consoles and smartphones"
`experience_games_vr`	Bool	Participant has experience in "Playing video games on a virtual reality (VR) headset"
`experience_360_interaction`	Bool	Participant has experience in "Interacting with a 360 degree video"
`experience_point_cloud_interaction`	Bool	Participant has experience in "Interacting with point clouds"
`experience_computer_graphics`	Bool	Participant has experience in "Computer graphics (a subarea of computer science)"
`first_object`	String	The first point cloud object presented to the participant (`dancer` or `thaidancer`)
`duration_total`	Int	Total time (in milliseconds) spent for the study, including time for feedback
`duration_introduction`	Int	Time (in milliseconds) spent in the introduction
`duration_reference_1`	Int	Time (in milliseconds) spent watching the first reference video (see `first_object`)
`duration_reference_2`	Int	Time (in milliseconds) spent watching the second reference video

Ratings

The table ratings.csv contains the quality ratings of each participant for each of the 120 experiment configurations:

Column Name	Data Type	Description
`participant`	Int	ID of the participant who submitted this rating
`object`	String	Shown object, is either `dancer` or `thaidancer`
`distance`	String	Distance of the camera to the object in `near` (2.5 units), `medium` (4.5 units), `far` (8.5 units)
`frame_rate`	Int	Frame rate in `10`, `15`, `30`
`encode_method`	String	Used encoding method, either `Draco` or (frame-by-frame) `V-PCC`
`quantization_level_index`	Int	Quantization level index from `0` (high compression, low quality) to `4` (low compression, high quality)
`qoe`	Int	Quality rating in `1` (bad), `2` (poor), `3` (fair), `4` (good), `5` (excellent) for this configuration
`duration`	Int	Time (in milliseconds) spent watching this experiment configuration
`size`	Int	Video size in the browser (attribute `height` of the square-shaped `video` HTML element)
`order`	Int	Ordering of the experiments for each participant from `0` to `119`

Resources

The table resources.csv contains encoding and decoding time together with the bit rate of the original and compressed streams. Note that the user study was restricted to object rate 30 for Draco.

Column Name	Data Type	Description
`object`	String	Object type, either `dancer` or ``thaidancer
`encode_method`	String	Used encoding method, either `Draco` or (frame-by-frame) `V-PCC`
`quantization_level_index`	Int	Quantization level index from `0` (high compression, low quality) to `4` (low compression, high quality)
`object_rate`	Int	Number of object per seconds in the sequence
`encoding_time`	Float	Encoding time per second of point cloud stream in seconds
`decoding_time`	Float	Decoding time per second of point cloud stream in seconds
`bit_rate`	Float	Bit rate of the uncompressed point cloud sequence in MBit/s
`compressed_bit_rate`	Float	Bit rate of the compressed point cloud sequence in MBit/s using the configuration from this row

Overview and QoE Models

This repository contains jupyter notebooks that provide an overview of the data (qoe_overview.ipynb) and create qoe models using various classical machine learning methods (qoe_model.ipynb). Although the models have a fixed random state, we noticed that the results can differ slightly between different systems.

The code has been tested with Python 3.11.3, the requirements are in requirements.txt.

Reference

You can find our paper here. If you use material provided in this repository, please cite it with

@INPROCEEDINGS{weil23pointCloudqoe,
  author={Weil, Jannis and Alkhalili, Yassin and Tahir, Anam and Gruczyk, Thomas and Meuser, Tobias and Mu, Mu and Koeppl, Heinz and Mauthe, Andreas},
  booktitle={15th International Conference on Quality of Multimedia Experience (QoMEX)}, 
  title={Modeling Quality of Experience for Compressed Point Cloud Sequences based on a Subjective Study}, 
  year={2023},
  volume={},
  number={},
  pages={135-140},
  doi={10.1109/QoMEX58391.2023.10178579}
}