REVEX: A Unified Framework for Removal-Based Explainable Artificial Intelligence in Video

TL;DR

REVEX introduces a comprehensive framework for removal-based video explanations, adapting existing techniques to include temporal data and evaluating their effectiveness across various networks and metrics.

Contribution

This work extends fine-grained explanation methods to video by incorporating temporal information and systematically evaluating their performance and limitations.

Findings

Video LIME outperformed others in deletion and insertion metrics.

Video RISE achieved the best in average drop metric.

Localization-based metrics showed low performance across methods.

Abstract

We developed REVEX, a removal-based video explanations framework. This work extends fine-grained explanation frameworks for computer vision data and adapts six existing techniques to video by adding temporal information and local explanations. The adapted methods were evaluated across networks, datasets, image classes, and evaluation metrics. By decomposing explanation into steps, strengths and weaknesses were revealed in the studied methods, for example, on pixel clustering and perturbations in the input. Video LIME outperformed other methods with deletion values up to 31\% lower and insertion up to 30\% higher, depending on method and network. Video RISE achieved superior performance in the average drop metric, with values 10\% lower. In contrast, localization-based metrics revealed low performance across all methods, with significant variation depending on network. Pointing game accuracy reached 53\%, and IoU-based metrics remained below 20\%. Drawing on the findings across XAI methods, we further examine the limitations of the employed XAI evaluation metrics and highlight their suitability in different applications.