Are Synthetic Corruptions A Reliable Proxy For Real-World Corruptions?

TL;DR

This study evaluates whether synthetic corruptions can reliably stand in for real-world corruptions in testing the robustness of semantic segmentation models, finding strong correlations that support their use.

Contribution

It provides the largest benchmarking comparison between synthetic and real-world corruptions for semantic segmentation, offering insights into their correlation and reliability.

Findings

Strong correlation in mean performance between synthetic and real-world corruptions

Analysis of corruption-specific correlations reveals when synthetic corruptions are effective

Supports using synthetic corruptions as a resource-efficient robustness testing proxy

Abstract

Deep learning (DL) models are widely used in real-world applications but remain vulnerable to distribution shifts, especially due to weather and lighting changes. Collecting diverse real-world data for testing the robustness of DL models is resource-intensive, making synthetic corruptions an attractive alternative for robustness testing. However, are synthetic corruptions a reliable proxy for real-world corruptions? To answer this, we conduct the largest benchmarking study on semantic segmentation models, comparing performance on real-world corruptions and synthetic corruptions datasets. Our results reveal a strong correlation in mean performance, supporting the use of synthetic corruptions for robustness evaluation. We further analyze corruption-specific correlations, providing key insights to understand when synthetic corruptions succeed in representing real-world corruptions. Open-source Code: https://github.com/shashankskagnihotri/benchmarking_robustness/tree/segmentation_david/semantic_segmentation