Empowering Networks With Scale and Rotation Equivariance Using A Similarity Convolution

Authors: Zikai Sun, Thierry Blu

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate the efficacy of our approach in the image classification task, demonstrating its robustness and the generalization ability to both scaled and rotated inputs.5 EXPERIMENTS
Researcher Affiliation Academia Zikai Sun and Thierry Blu Department of Electronic Engineering, The Chinese University of Hong Kong zksun@link.cuhk.edu.hk, thierry.blu@m4x.org
Pseudocode Yes Algorithm 1: Pipeline of our SREN Algorithm
Open Source Code No The paper does not contain any explicit statement about releasing open-source code for the described methodology, nor does it provide a link to a code repository.
Open Datasets Yes The MNIST-ROT-12K DATASET (Larochelle et al., 2007) is commonly used to evaluate rotation-equivariant algorithms.To evaluate the generalization ability of our method, we conduct experiments on the STL-10 dataset Coates et al. (2011).
Dataset Splits No Specifically, we pad the images to 56 56 pixels, keep the training set unchanged, and randomly apply rotations, scalings, and translations to each test image within the ranges of θ = [0, 2π), s = [1, 2[, and t = 10.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU models, or memory specifications used for running the experiments.
Software Dependencies No The paper mentions using Adam optimizer but does not provide specific version numbers for any software dependencies or libraries.
Experiment Setup Yes We use Adam optimizer (Kingma & Ba, 2015) with a weight decay of 0.01, initialize the weights with Xavier (Glorot & Bengio, 2010), and set the learning rate to 0.01, which decays by a factor of 0.1 every 50 epochs. We set the batch size to 128 and stop training after 200 epochs.The network is trained for 1000 epochs with a batch size of 128, using Adam as the optimizer. The initial learning rate is set to 0.1 and adjusted using a cosine annealing schedule during training.