Visual Results by Method

Results for methods appear here after users upload them and approve them for public display.





Metrics


Average error
Error visualization: for each pair, the plot on the left reports the errors in cm for all the ground-truth correspondences (sorted by decreasing error).

For the intra-subject challenge, the figure on the right shows the error distribution over the vertices of the first scan in the pair (blue represents the minimum error, red represents the maximum error).

For the inter-subject challenge, the figure on the right shows the error distribution over different areas of the first scan in the pair (each area corresponds to a landmark location; black represents the minimum error, white represents the maximum error).


References
[1]
A
Anonymous.
[2]
CHARM
[3]
Convex-Opt
Robust Nonrigid Registration by Convex Optimization. Qifeng Chen, Vladlen Koltun. International Conference on Computer Vision (ICCV), 2015
[4]
FMNet
"Deep functional maps: Structured prediction for dense shape correspondence". Litany, Remez, Rodola, Bronstein, Bronstein. Proc. ICCV 2017
[5]
Inter Lap baseline 80-60
Anonymous.
[6]
Inter Lap baseline 80-60
Anonymous.
[7]
Inter-L-xyz
Anonymous.
[8]
Intra L
Anonymous.
[9]
Intra L
Anonymous.
[10]
Intra L-refined
Anonymous.
[11]
Intra-A-xyz
Anonymous.
[12]
Intra-L-xyz
Anonymous.
[13]
Intra-M
Anonymous.
[14]
MLP baseline
Anonymous. baseline model using only MLP. And first 80 models 60 epochs
[15]
Shape correspondences from learnt template-based parametrization
Shape correspondences from learnt template-based parametrization, Groueix, Thibault and Fisher, Matthew and Kim, Vladimir G. and Russell, Bryan and Aubry, Mathieu, ECCV 2018.
[16]
SP_intra_challenge
S. Zuffi, M. J. Black, "The Stitched Puppet: A Graphical Model of 3D Human Shape and Pose", CVPR, Boston, MA, June 2015.
[17]
test_avg
Anonymous. use first 80 models of the FAUST. Train 60 epoch.