Stereopagnosia: Fooling Stereo Networks with Adversarial Perturbations

Authors: Alex Wong, Mukund Mundhra, Stefano Soatto2879-2888

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test our method using the most recent stereo networks and evaluate their performance on public benchmark datasets.
Researcher Affiliation Academia Alex Wong:, Mukund Mundhra:, Stefano Soatto UCLA Vision Lab alexw@cs.ucla.edu, mukundmundhra@cs.ucla.edu, soatto@cs.ucla.edu
Pseudocode No The paper describes the methods using mathematical equations and text, but does not provide structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide a statement or link to the open-source code for the described methodology.
Open Datasets Yes We test our method using the most recent stereo networks and evaluate their performance on public benchmark datasets. ... KITTI 2015 stereo (Menze and Geiger 2015) validation set in the main paper and KITTI 2012 (Geiger, Lenz, and Urtasun 2012) validation set in the Supp. Mat.
Dataset Splits Yes Following KITTI validation protocol, the KITTI 2015 training set is divided into 160 for training and 40 for validation, and the KITTI 2012 training set is split into 160 for training and 34 for validation.
Hardware Specification No No specific details about the hardware (e.g., GPU models, CPU types, memory) used for running the experiments are provided in the paper.
Software Dependencies No No specific software dependencies with version numbers (e.g., programming languages, libraries, frameworks) are provided in the paper.
Experiment Setup Yes We study perturbations under four different upper norms, ϵ t0.02, 0.01, 0.005, 0.002u. ... When optimizing with I-FGSM and DI2-FGSM, we used N 40 and α 1{N ϵ for ϵ t0.01, 0.005, 0.002u and α 0.10ϵ for ϵ 0.02. For MI-FGSM and MDI2-FGSM, α 1{N ϵ for all ϵ and chose β 0.47 for momentum.