Submission

Important dates

Results submission: February 17th, 2017 (passed)

Paper submission: February 28th 2017 (passed)

Format

Participants are required to submit a distance matrix for each retrieval setting (one or both), i.e., a dissimilarity score must be evaluated between every pair of shapes. In particular:

  • Element D(i,j) of the matrix D must contain a value in [0,1] indicating the dissimilarity/distance between shape i and shape j
  • Use -1 if your method could not process/evaluate the pair of shapes (due to long runtimes, crash, etc.)
  • Shapes are indexed according to their filenames, e.g., i=32 refers to shape32.off
  • Matrix D is not strictly required to be symmetric
  • Write the matrix to file in single-precision binary format; this should result in a few MB per matrix. In Matlab, you can set D=single(D) and use save('D.mat','D','-mat'). If you do not use Matlab, please write a binary file in row-major order, with 4 bytes per similarity value.

Up to 3 submissions (resulting from different runs or parameter settings) are allowed for each participating method. Please also provide indicative per-query runtimes and the hardware specs you used for the experiments.

In your email, please indicate the name of your method(s). You will be asked, at a later time, a brief description (~1 page and 1 figure) to be included in the final paper.

Evaluation

Each method will be evaluated based on its distance matrices, using standard retrieval scores (nearest neighbor, first tier, second tier, etc.). For the evaluation, we are adopting code from the SHREC 2015 track on Non-rigid 3D Shape Retrieval.