Detection Transformers Under the Knife: A Neuroscience-Inspired Approach to Ablations

TL;DR

This paper systematically ablates key components of detection transformers inspired by neuroscience studies, revealing their roles in model performance and suggesting avenues for simplification and improved transparency.

Contribution

It introduces a neuroscience-inspired ablation methodology for detection transformers and provides detailed insights into component importance and redundancy.

Findings

DETR is sensitive to encoder MHSA and decoder MHCA ablations

DDETR's deformable attention enhances robustness

DINO's look-forward update rule increases resilience

Abstract

In recent years, Explainable AI has gained traction as an approach to enhancing model interpretability and transparency, particularly in complex models such as detection transformers. Despite rapid advancements, a substantial research gap remains in understanding the distinct roles of internal components - knowledge that is essential for improving transparency and efficiency. Inspired by neuroscientific ablation studies, which investigate the functions of brain regions through selective impairment, we systematically analyze the impact of ablating key components in three state-of-the-art detection transformer models: Detection transformer (DETR), deformable detection transformer (DDETR), and DETR with improved denoising anchor boxes (DINO). The ablations target query embeddings, encoder and decoder multi-head self-attentions (MHSA) as well as decoder multi-head cross-attention (MHCA) layers. We evaluate the effects of these ablations on the performance metrics gIoU and F1-score, quantifying effects on both the classification and regression sub-tasks on the COCO dataset. To facilitate reproducibility and future research, we publicly release the DeepDissect library. Our findings reveal model-specific resilience patterns: while DETR is particularly sensitive to ablations in encoder MHSA and decoder MHCA, DDETR's multi-scale deformable attention enhances robustness, and DINO exhibits the greatest resilience due to its look-forward twice update rule, which helps distributing knowledge across blocks. These insights also expose structural redundancies, particularly in DDETR's and DINO's decoder MHCA layers, highlighting opportunities for model simplification without sacrificing performance. This study advances XAI for DETRs by clarifying the contributions of internal components to model performance, offering insights to optimize and improve transparency and efficiency in critical applications.