THE CONTEST IS NOW OVER! The SHREC'2012 track on 3D-mesh segmentation is now over. Results will be published during the 3DOR'2012 worshop.
Unlike the previous SHREC contests, the objective of this SHREC 2012 contest is to evaluate the performance of 3D-mesh segmentation techniques instead of evaluating the performance of 3D-mesh retrieval techniques. The segmentation process can be used as a pre-processing step to multiple applications including the retrieval task.
The task of the participants is to segment the models of our dataset, like human observers do (i.e. segmentation of the model into meaningful parts). To evaluate the algorithms , we will rely on the ground-truth corpus and metrics defined is this recent work:
The figure below shows one example of 3D model from the dataset and a ground-truth manual segmentation.
Obviously, the ground-thruth collection and the on-line benchmark (described in the above cited paper) will be unavaillable for the duration of the contest.
The dataset is composed of 28 watertight models, grouped in five classes: animal, furniture, hand, human and bust.
The dataset is available as a single ZIP file containing the 28 3D-meshes in three different formats: VRML97 (modelname.wrl), PLY (modelname.ply) and OFF (modelname.off).
Click here to download the ZIP file (25 MB).
For the evaluation, each of these models will be associated with 4 ground-truth segmentations done by human subjects.
The evaluation protocol is the following: for each model, the segmentation from a given algorithm is compared to the ground truth segmentations. Hence a quality score is obtained for each model and each algorithm based on the 3D-NPRI metric (see the paper cited above).
Participants must run their segmentation algorithms on the models of the dataset and produce segmentations in the LAB file format. The LAB format is a simple format which contains the following information: number of vertices and number of regions on the first line, then for each vertex, label of the region to which it belongs.
Here is a LAB file example corresponding to a mesh with 5 vertices, segmented in 2 regions. The first three vertices are belonging to the segment labeled 1, and the last two vertices to the segment labeled 2:
5 2 1 1 1 2 2
Each segmented mesh must be stored in a file with the same name as the original mesh but in the LAB format. For instance, the segmentation corresponding to the model bunny.wrl must respect the LAB file format and be named bunny.lab. Finally the 28 resulting LAB files (ie. 28 segmentations corresponding to the 28 meshes of the dataset) have to be stored in a single ZIP file.
Here are the different steps to follow by the participants:
January 24, 2012 — call for participation;
February 3, 2012 February 10, 2012 — deadline for registration by email;
February 10, 2012February 15, 2012 — deadline for result submission by email;
February 14, 2012February 21, 2012 — return evaluation results;
February 17, 2012February 24, 2012 — deadline for 1-page summary paper.
The organizers of this track are:
To contact them all, please use the following email address: liste-shrec12-segmentation at telecom-lille1 dot eu.
Last update: 2012-02-29