Deformation-Driven Shape Correspondence via Shape Recognition

Chenyang Zhu1,2,  Renjiao Yi1,2,  Wallace Lira1, Ibraheem Alhashim1,  Kai Xu2, Hao Zhang1

1Simon Fraser University,     2National University of Defense Technology

ACM Transactions on Graphics (SIGGRAPH 2017), 36(4)


An overview of our deformation-driven correspondence algorithm. The input consists of two pre-segmented or over-segmented 3D shapes. We recursively split and match substructures according to a data-driven plausibility criterion that relies exclusively on shape recognition. The first iteration splits the input shapes into two sub-shapes. Given a pair of matched components, the algorithm recursively splits and matches them. Finally, aer termination conditions are met for each substructure, we obtain a final part correspondence.


Many approaches to shape comparison and recognition start by establishing a shape correspondence. We “turn the table” and show that quality shape correspondences can be obtained by performing many shape recognition tasks. What is more, the method we develop computes a fine-grained, topology-varying part correspondence between two 3D shapes where the core evaluation mechanism only recognizes shapes globally. This is made possible by casting the part correspondence problem in a deformation-driven framework and relying on a data-driven “deformation energy” which rates visual similarity between deformed shapes and models from a shape repository. Our basic premise is that if a correspondence between two chairs (or airplanes, bicycles, etc.) is correct, then a reasonable deformation between the two chairs anchored on the correspondence ought to produce plausible, “chair-like” in-between shapes.

Given two 3D shapes belonging to the same category, we perform a top-down, hierarchical search for part correspondences. For a candidate correspondence at each level of the search hierarchy, we deform one input shape into the other, while respecting the correspondence, and rate the correspondence based on how well the resulting deformed shapes resemble other shapes from ShapeNet belonging to the same category as the inputs. The resemblance, i.e., plausibility, is measured by comparing multi-view depth images over category-specic features learned for the various shape categories. We demonstrate clear improvements over state-of-the-art approaches through tests covering extensive sets of man-made models with rich geometric and topological variations.

Paper PDF


PDF, 30.9MB

Algorithm Overview

Left: Samples of our plausibility measure training data. "Missing negative" (left-middle) samples shows examples with patches that are not present in the shape, while "swap negative" (left-right) shows structures that have some undesirable combination of patches. Either negative case directly affects the visual perception of shape implausibility. Right: Extracting a feature vector from the HOG representation of the depth image and the mid-level patches. Feature vectors are constructed by generating a response map using a convolution operation between patches and depth maps. We obtain a 10-dimensional feature vector for each depth patch by averaging five segments of the response map in the vertical and horizontal directions. The feature vectors of all the patches of a given depth image are then combined into one single feature vector that represents the depth image.


Some hierarchical correspondence of input shape segments obtained by our method.


A gallery of part correspondences computed by our algorithm (boom pair of each two shape pairs compared) with comparisons to GeoTopo [Alhashim et al. 2015]. Matched parts share the same color; unmatched parts are in gray.

We would like to thank the reviewers for their valuable comments and feedback. We also thank Noa Fish and Oliver van Kaick for fruitful discussions. This work is supported in part by grants from NSERC Canada (611770), China Scholarship Council, and NSF China (61572507, 61532003, 61622212).


Demo code with dataset.

@article {zhu_sig17,
    title = {Deformation-Driven Shape Correspondence via Shape Recognition},
    author = {Chenyang Zhu and Renjiao Yi and Wallace Lira and Ibraheem Alhashim and Kai Xu and Hao Zhang},
    journal = {ACM Transactions on Graphics (Proc. of SIGGRAPH 2017)},
    volume = {36},
    number = {4},
    pages = {to appear},
year = {2017}


    Back to TopHome