Is In-Domain Data Really Needed? A Pilot Study on Cross-Domain Calibration for Network Quantization

TL;DR

This study investigates whether out-of-domain data can effectively calibrate neural networks for quantization, demonstrating stable performance across diverse image domains and identifying domain similarity as a key factor.

Contribution

The paper shows that cross-domain calibration is viable for network quantization and introduces Gram matrix similarity as a criterion for selecting calibration data.

Findings

Cross-domain calibration maintains stable quantization performance.

Performance correlates with Gram matrix similarity between source and calibration domains.

Out-of-domain calibration enables privacy-preserving model deployment.

Abstract

Post-training quantization methods use a set of calibration data to compute quantization ranges for network parameters and activations. The calibration data usually comes from the training dataset which could be inaccessible due to sensitivity of the data. In this work, we want to study such a problem: can we use out-of-domain data to calibrate the trained networks without knowledge of the original dataset? Specifically, we go beyond the domain of natural images to include drastically different domains such as X-ray images, satellite images and ultrasound images. We find cross-domain calibration leads to surprisingly stable performance of quantized models on 10 tasks in different image domains with 13 different calibration datasets. We also find that the performance of quantized models is correlated with the similarity of the Gram matrices between the source and calibration domains, which can be used as a criterion to choose calibration set for better performance. We believe our research opens the door to borrow cross-domain knowledge for network quantization and compression.