Correspondence learning via linearly-invariant embedding

Authors: Riccardo Marin, Marie-Julie Rakotosaona, Simone Melzi, Maks Ovsjanikov

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our pipeline on the correspondence problem between non-rigid 3D point clouds in the challenging class of human models. Finally, we demonstrate that our approach achieves state-of-the-art results in challenging non-rigid 3D point cloud correspondence applications.
Researcher Affiliation Academia Riccardo Marin University of Verona riccardo.marin_01@univr.it Marie-Julie Rakotosaona LIX, Ecole Polytechnique, IP Paris mrakotos@lix.polytechnique.fr Simone Melzi LIX, Ecole Polytechnique, IP Paris Sapienza University of Rome melzi@di.uniroma1.it Maks Ovsjanikov LIX, Ecole Polytechnique, IP Paris maks@lix.polytechnique.fr
Pseudocode No The paper describes the pipeline and methods in text and uses flow diagrams (Figure 1), but no explicit pseudocode or algorithm blocks are provided.
Open Source Code Yes The code, datasets and our pre-trained networks can be found online: https://github.com/riccardomarin/Diff-FMaps.
Open Datasets Yes For our experiments we train over 10K shapes from the SURREAL dataset [57], resampled at 1K vertices. We consider a first test set composed by the 100 shapes from the FAUST dataset [6] (10 different subjects in 10 different poses). [1] Scan the world project. https://www.myminifactory.com/scantheworld
Dataset Splits No The paper mentions training on the SURREAL dataset and testing on the FAUST dataset but does not provide specific train/validation/test split percentages or sample counts for reproducibility.
Hardware Specification No The paper does not provide any specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments.
Software Dependencies No The paper states that 'Both of our networks N and G are built upon the Point Net architecture [44]' but does not provide specific version numbers for any software dependencies, libraries, or programming languages used in its implementation.
Experiment Setup Yes Both of our networks N and G are built upon the Point Net architecture [44]. For our experiments we train over 10K shapes from the SURREAL dataset [57], resampled at 1K vertices. We learn a k = 20 dimensional embedding (basis) and p = 40 probe functions for each point cloud.