CaSPR: Learning Canonical Spatiotemporal Point Cloud Representations

Authors: Davis Rempe, Tolga Birdal, Yongheng Zhao, Zan Gojcic, Srinath Sridhar, Leonidas J. Guibas

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

Reproducibility Variable Result LLM Response
Research Type Experimental We now evaluate the canonicalization, representation, and reconstruction capabilities of Ca SPR, demonstrate its utility in multiple downstream tasks, and justify design choices. Dataset and Preprocessing: We introduce a new dataset containing simulated rigid motion of objects in three Shape Net [5] categories: cars, chairs, and airplanes. ... Evaluation Procedure: To measure canonicalization errors, T-NOCS coordinates are split into the spatial and temporal part with GT given by X and t respectively. The spatial error at frame k is 1 Mk PMk i=1 bxi xi 2 and the temporal error is 1 Mk PMk i=1 |bti ti| . For reconstruction, the Chamfer Distance (CD) and Earth Mover s Distances (EMD) are measured (and reported multiplied by 103). Lower is better for all metrics; we report the median over all test frames because outlier shapes cause less informative mean errors.
Researcher Affiliation Academia 1Stanford University 2University of Padova 3ETH Zürich
Pseudocode No The paper describes the method and architecture in text and diagrams but does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper provides a project website URL (geometry.stanford.edu/projects/caspr) but does not explicitly state that the source code for the methodology is available there, nor does it provide a direct link to a code repository.
Open Datasets Yes We introduce a new dataset containing simulated rigid motion of objects in three Shape Net [5] categories: cars, chairs, and airplanes.
Dataset Splits No The paper describes training and testing data usage ('For training, 5 frames...At test time, we use a different...'), but does not explicitly mention or specify details for a separate validation dataset split.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory, or cloud instances) used for running its experiments.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies used in the experiments.
Experiment Setup No The paper describes the loss functions and data sampling strategy but does not provide concrete hyperparameter values (e.g., learning rate, batch size, number of epochs) or system-level training settings in its main text.