Infrared Domain Adaptation with Zero-Shot Quantization

TL;DR

This paper presents a zero-shot quantization method for infrared imaging models that uses batch normalization statistics to calibrate models without training data, demonstrating effectiveness in object detection tasks.

Contribution

It introduces a zero-shot quantization framework tailored for infrared domain models, leveraging batch normalization statistics for calibration without training data.

Findings

Zero-shot quantization effectively calibrates infrared models without training data.

The method compares favorably with post-training quantization on thermal datasets.

Batch normalization statistics are crucial for successful zero-shot quantization.

Abstract

Quantization is one of the most popular techniques for reducing computation time and shrinking model size. However, ensuring the accuracy of quantized models typically involves calibration using training data, which may be inaccessible due to privacy concerns. In such cases, zero-shot quantization, a technique that relies on pretrained models and statistical information without the need for specific training data, becomes valuable. Exploring zero-shot quantization in the infrared domain is important due to the prevalence of infrared imaging in sensitive fields like medical and security applications. In this work, we demonstrate how to apply zero-shot quantization to an object detection model retrained with thermal imagery. We use batch normalization statistics of the model to distill data for calibration. RGB image-trained models and thermal image-trained models are compared in the context of zero-shot quantization. Our investigation focuses on the contributions of mean and standard deviation statistics to zero-shot quantization performance. Additionally, we compare zero-shot quantization with post-training quantization on a thermal dataset. We demonstrated that zero-shot quantization successfully generates data that represents the training dataset for the quantization of object detection models. Our results indicate that our zero-shot quantization framework is effective in the absence of training data and is well-suited for the infrared domain.