HQOD: Harmonious Quantization for Object Detection

TL;DR

This paper introduces HQOD, a framework that improves object detection accuracy by addressing task inharmony issues during quantization-aware training, achieving state-of-the-art results on MS COCO.

Contribution

The paper proposes a novel task-correlated loss and a harmonious IoU loss to enhance quantized object detectors, effectively mitigating task inharmony and performance degradation.

Findings

Achieves 39.6% mAP with 4-bit quantization on MS COCO

Outperforms full-precision detectors with the proposed method

Easily integrates into various QAT algorithms and detectors

Abstract

Task inharmony problem commonly occurs in modern object detectors, leading to inconsistent qualities between classification and regression tasks. The predicted boxes with high classification scores but poor localization positions or low classification scores but accurate localization positions will worsen the performance of detectors after Non-Maximum Suppression. Furthermore, when object detectors collaborate with Quantization-Aware Training (QAT), we observe that the task inharmony problem will be further exacerbated, which is considered one of the main causes of the performance degradation of quantized detectors. To tackle this issue, we propose the Harmonious Quantization for Object Detection (HQOD) framework, which consists of two components. Firstly, we propose a task-correlated loss to encourage detectors to focus on improving samples with lower task harmony quality during QAT. Secondly, a harmonious Intersection over Union (IoU) loss is incorporated to balance the optimization of the regression branch across different IoU levels. The proposed HQOD can be easily integrated into different QAT algorithms and detectors. Remarkably, on the MS COCO dataset, our 4-bit ATSS with ResNet-50 backbone achieves a state-of-the-art mAP of 39.6%, even surpassing the full-precision one.