Log-Polar Space Convolution Layers

Authors: Bing Su, Ji-Rong Wen

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on different tasks and datasets demonstrate the effectiveness of the proposed LPSC.
Researcher Affiliation Academia Bing Su, Ji-Rong Wen Beijing Key Laboratory of Big Data Management and Analysis Methods Gaoling School of Artificial Intelligence, Renmin University of China Beijing 100872, China subingats@gmail.com; jrwen@ruc.edu.cn
Pseudocode No The paper describes the calculation of LPSC using mathematical equations (Eq. 1, Eq. 3) and descriptive text, but it does not include a formally labeled 'Pseudocode' or 'Algorithm' block.
Open Source Code Yes Our code is available at https://github.com/Bing Su12/ Log-Polar-Space-Convolution.
Open Datasets Yes For image classification, we evaluate the behaviors of LPSC integrated with different CNN architectures on three datasets: CIFAR-10, CIFAR-100 [52], and Image Net [53].
Dataset Splits Yes Image Net [53] contains 1.28 million training images and 50k validation images from 1000 classes.
Hardware Specification No The paper mentions 'Due to the limitation of computing resources, we reduced the batch size and learning rate by 4 times,' but does not provide specific hardware details such as GPU models, CPU types, or cloud instance specifications used for the experiments.
Software Dependencies No We use the Pytorch [54] implementation2 of these architectures as our baseline.
Experiment Setup Yes To make a fair comparison, all experimental setup and details including the learning rate, batch size, number of filters per layer, hyper-parameters for the optimizer (e.g., γ, momentum, weight decay) remain exactly the same as in the baseline.