3DCOMET: 3D Compression Methods Test Dataset

Tools

The 3DCOMET dataset is released in conjunction with two different tools for computing results statistics. These tools can be downloaded from:
https://bitbucket.org/vmorell/3dcomet-tools/

Compression tester

The first one is a compression tester appllication, which analyses the results of given 3D compression methods. It starts from a directory with three subdirectories: original, compressed and decompressed data. These three subdirectories must keep the original structure. The original data are the images in binary (uncompressed) format. Then, from the compressed data, each file is compared with the original one, obtaining the compression ratio. Some metrics are calculated, which will be used later in conjunction with the second tool to generate some graphs about the obtained results. This application has been developed in C++ using the PointCloud library
Usage:

compressionTester general_path original_folder compressed_folder reconstructed_folder

Statistics computation

The next tool uses a file with the statistics obtained from the dataset: curvature and entropy. In the GitHub repositorie, there is a project (computeStats) which can be used for generating those statistics. This project generates the statistics for one file:

computeStats pointCloud

The statistics for all the dataset have been generated in a single file: statistics.csv. Using the output from the compression tester and this file, the both could be concatenate in one single file, which is the one used for the next tool.

concatStats compressionTesterOutput statistics.csv > jointStatistics

Statistics generator

The second tool is a Gnuplot script that provides with the following capabilities:

Results comparison: this tool generates three graphs that compare the results between different compression methods. It takes as an input the data generated by the previous utility provided in this dataset REF. The first graph shows the compression rate obtained for each compression method. The second graph shows the distance error for each uncompressed 3D point regard the original one. Finally, the third graph shows the color error using the nearest neighbour (3D point) from the original point cloud.

In conjunction with the dataset, five methods are provided as a baseline for further progresses on the development of new methods for 3D data compression. In ProvidedResults.zip, the statistics for these five methods are provided and they can serve as a baseline for future comparison.

gnuplot -e "f1='Octree24'" -e "f2='PCD'" -e "f3='LZ77'" -e "f4='Burrows'" -e "f5='Morell2014k1'" -e "f6='Morell2014k5'"  -e "m='SY TE'"  -e "t='pdf'" plot.gp

A different number of files can be passed as an argument. Each file must be separated using the the following option:

-e "fi='name'"

The Gnuplot script allows to compare up to 6 files, each file must be specified with the option "fi='name'" where i is the number of the file. Moreover, i must start in 1 and has to be sequentially increase.
In addition, the Gnuplot script allows to specify the output by using the option -e "t='terminal'" where the kind of terminal must be specified. The default terminal is the one present in your operating system. You can use the option 'pdf' to export the output to a PDF file. For generating LaTeX content you can use the 'latex' option. Examples of terminals are x11 for Linux or aqua or qt for Mac.

-e "t='aqua'"

We can use the option -e "m='SY TE'" to specify if the output must compute statistics information for a subset of the whole dataset. The first part specifies if the (CO)mplete, (SY)nthetic or the (RE)al dataset is used and the second one specifies if the script only has to compute the results depending on (TE)xture or (ST)ructure. Example:

-e "m='SY TE'"

Below are shown the different graphs generated using the Gnuplot script and using the four baseline methods for 3D point cloud compression provided with the dataset. These graphs show the compression rate (%), the distance error and the texture error.

Finally, a detail plot can be obtained, using two quantitative metrics. For structure, the curvature of a pointcloud is used and, for texture, the entropy. To select this option, we need to provide the following option to the command:

-e "l='true'"

In this way, the graphs for compression rate, distance error and color error with respect to curvature and entropy for real pointclouds are presented. If we need to select synthetic images, an additional option must be given: