Deep Functional Dictionaries: Learning Consistent Semantic Structures on 3D Models from Functions

Authors: Minhyuk Sung, Hao Su, Ronald Yu, Leonidas J. Guibas

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness of our technique in various segmentation and keypoint selection applications. In experiments, we test our model with existing neural network architectures, and demonstrate the performance on labeled/unlabeled segmentation and keypoint correspondence problem in various datasets.
Researcher Affiliation Academia Minhyuk Sung Stanford University mhsung@cs.stanford.edu Hao Su University of California San Diego haosu@eng.ucsd.edu Ronald Yu University of California San Diego ronaldyu@ucsd.edu Leonidas Guibas Stanford University guibas@cs.stanford.edu
Pseudocode Yes 1: function SINGLE STEP GRADIENT ITERATION(X, f, Θt, η) 2: Compute: At = A(X; Θt). 3: Solve: xt = arg minx Atx f 2 2 s.t. C(x). 4: Update: Θt+1 = Θt η L(A(X; Θt); f, xt). 5: end function Algorithm 1: Single-Step Gradient Iteration.
Open Source Code Yes Code for all experiments below is available in https://github.com/mhsung/deep-functional-dictionaries.
Open Datasets Yes Yi et al. [41] provide keypoint annotations on 6,243 chair models in Shape Net [7]. Stanford 3D Indoor Semantic Dataset (S3DIS) [2] is a collection of real scan data of indoor scenes with annotations of instance segments and their semantic labels. Here, we test with 100 non-rigid human body shapes in MPI-FAUST dataset [4].
Dataset Splits Yes We follow their experiment setup by using the same split of training/validation/test sets and the same 2k sampled point cloud as inputs.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions using Point Net [32] architecture but does not specify software versions for libraries, frameworks, or operating systems.
Experiment Setup Yes In the experiment, we use a 80-20 random split for training/test sets 1, train the network with 2k point clouds as provided by [41], and set k = 10 and γ = 0.0. Table 1 shows the results of our method when using k = 10 and γ = 1.0. Thus, we use k = 150 and γ = 1.0. σ is Gaussian-weighting parameter (0.001 in our experiment).