Large-scale Point Cloud Quality Assessment Dataset (LS-PCQA)

Yipeng Liu


The established large-scale PCQA dataset named LS-PCQA includes 104 reference point clouds and more than 22,000 distorted samples. In the dataset, each reference point cloud is augmented with 31 types of impairments at 7 distortion levels. Besides, each distorted point cloud is assigned with a pseudo quality score as its substitute of Mean Opinion Score (MOS).

Reference Point Clouds

104 different point clouds include 28 human body models, 48 animal body models and 28 inanimate objects. The following image illustrates the snapshots of all the reference point clouds for LS-PCQA.


In total, 31 distortion types are considered.

Distortion TypesDistortion Types
Additive Gaussian noisePoisson noise
Color noiseGaussian geometry shifting
High frequency noiseUniform geometry shifting
Quantization noiseLocal missing
Mean shift (intensity shift)Local offset
Contrast changeLocal rotation
Change of color saturationLuminance noise
Spatially correlated noisePoisson Reconstruction
Multiplicative Gaussian noiseGPCC-lossless G and lossy A
Color quantization with ditherGPCC-lossless G and nearlossless A
Octree CompressionGPCC-lossy G and lossy A
Down sampleVPCC-lossy G and lossy A
Saltpepper noiseAVS-limitlossy G and lossy A
Rayleigh noiseAVS-lossless G and limitlossy A
Gamma noiseAVS-lossless G and lossy A
Uniform noise (white noise) 


Link for reference point clouds: BaiduNetDisk OneDrive

For a quick test, we supply 930 distorted samples with accurate MOS.

Link: BaiduNetDisk OneDrive

Link for whole dataset with generated pseudo MOS: BaiduNetDisk OneDrive