SERC Subjective Quality LiDAR Image Dataset

The SERC Subjective Quality LiDAR Image Dataset contains 342 reconstructed images and their ground-truth equivalents, as well as the MOS values and statistics obtained during subjective experiments conducted within the joint IMPRESS-U project entitled “EAGER IMPRESS-U: Exploratory Research on Generative Compression for Compressive Lidar”, funded in part by US National Science Foundation NSF under Grant No. 2404740, Science Technology Center in Ukraine (STCU) Agreement No. 7116, and National Science Centre, Poland (NCN), Grant no. 2023/05/Y/ST6/00197.
The dataset contains three subsets of 57 reconstructed pseudo-colored images and three subsets of their grayscale equivalents. The reconstructed images were prepared from HyperHeight Data Cubes using the developed method based on Generative Diffusion Models that modifies the probability flow ordinary differential equation to incorporate likelihood information approximated using the Kullback-Leibler divergence.
As the source data, the raw LiDAR data regarding the Smithsonian Environmental Research Center (SERC) in the state of Maryland (Latitude 38.88° N, Longitude 76.56° W) were used, which are available at the NEON (National Ecological Observatory Network) website: Discrete return LiDAR point cloud (DP1.30003.001), RELEASE-2024. (https://doi.org/10.48443/hj77-kf64). The reference CHMs with their reconstructions (57 pairs) were presented to several volunteer observers, and the reconstructed images were assessed on a 5-point scale in two independent experiments for both groups of pseudo-colored and grayscale images. The same procedure was conducted for the other subsets representing the 50th percentile and the DTM profiles (2nd percentile), also for pseudo-colored and grayscale images.
As part of a preliminary study of the visual quality of LiDAR images, 57 pairs of images were selected for eachsubset, consisting of original fragments of 192×192 pixels in size and the corresponding images obtained by the reconstruction algorithm. A subjective assessment of visual quality, defined as the degree of visual difference of the processed image relative to the original on a scale from 1 (terrible) to 5 (excellent, without noticeable or significant interference), was conducted by volunteer observers using the designed software. The methodology for conducting a subjective experiment consists of a relative comparison of pairs of images and determining the intensity of differences on a 5-point scale. Under the test conditions, participants have 10 seconds to determine the assessment based on their perception.
The order of MOS values in the provided text files, separately for pseudo-colored and grayscale images, is the same as in CSV files, where the filenames and statistics are provided as well.