Scalable Certified Segmentation via Randomized Smoothing

Authors: Marc Fischer, Maximilian Baader, Martin Vechev

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental evaluation on synthetic data and challenging datasets, such as Pascal Context, Cityscapes, and Shape Net, shows that our algorithm can achieve, for the first time, competitive accuracy and certification guarantees on real-world segmentation tasks.
Researcher Affiliation Academia 1Department of Computer Science, ETH Zurich, Switzerland. Correspondence to: Marc Fischer <marc.fischer@inf.ethz.ch>.
Pseudocode Yes Algorithm 1 adapted from (Cohen et al., 2019)
Open Source Code Yes We provide an implementation at https://github.com/ eth-sri/segmentation-smoothing.
Open Datasets Yes Our experimental evaluation on synthetic data and challenging datasets, such as Pascal Context, Cityscapes, and Shape Net, shows that our algorithm can achieve, for the first time, competitive accuracy and certification guarantees on real-world segmentation tasks.
Dataset Splits No The paper mentions using datasets like Cityscapes, Pascal Context, and ShapeNet, and refers to 'standard split of the dataset where available' in Appendix A.1, but it does not provide specific percentages, sample counts, or explicit citations to predefined splits in the main text or the relevant appendix section needed to reproduce the data partitioning.
Hardware Specification Yes All timings are for a single Nvidia Ge Force RTX 2080 Ti.
Software Dependencies Yes We used Python 3.8 and PyTorch 1.9 with CUDA 11.1 for all experiments.
Experiment Setup Yes All SEGCERTIFY (n0 = 10, α = 0.001) results are certifiably robust at radius R w.h.p. SEGCERTIFY n = 100, τ = 0.75 We trained our base models with Gaussian Noise (σ = 0.25), as in Cohen et al. (2019).