Spin-Weighted Spherical CNNs
Authors: Carlos Esteves, Ameesh Makadia, Kostas Daniilidis
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments show that our method outperforms previous methods on tasks like classification of spherical images, classification of 3D shapes and semantic segmentation of spherical panoramas. Every model is trained with different random seeds five times and averages and standard deviations (within parenthesis) are reported. See the appendix for training procedure details. |
| Researcher Affiliation | Collaboration | Carlos Esteves GRASP Laboratory University of Pennsylvania machc@seas.upenn.edu Ameesh Makadia Google Research makadia@google.com Kostas Daniilidis GRASP Laboratory University of Pennsylvania kostas@cis.upenn.edu |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | We will make our code and datasets publicly available at https://github.com/daniilidis-group/swscnn. |
| Open Datasets | Yes | We will make our code and datasets publicly available at https://github.com/daniilidis-group/swscnn. Our first experiment is on the Spherical MNIST dataset introduced by Cohen et al. [11]. We start from MNIST [31]. We tackle 3D object classification on Model Net40 [46]. We evaluate our method on the Stanford 2D3DS dataset [2]. |
| Dataset Splits | Yes | Three modes are evaluated depending on whether the training/test set are rotated (R) or not (NR). We follow Larochelle et al. [30] and swap the train and test sets so there are 10 k images for training and 50 k for test. Every model is trained with different random seeds five times and averages and standard deviations (within parenthesis) are reported. See the appendix for training procedure details. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments. |
| Software Dependencies | No | The paper mentions 'We follow Huffenberger and Wandelt [22] for the spin-weighted spherical Fourier transform (SWSFT) implementation', but it does not specify any software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x). |
| Experiment Setup | Yes | Every model is trained with different random seeds five times and averages and standard deviations (within parenthesis) are reported. See the appendix for training procedure details. |