Ef-QuantFace: Streamlined Face Recognition with Small Data and Low-Bit Precision

TL;DR

This paper introduces Ef-QuantFace, a face recognition model that achieves high accuracy with low-bit quantization using only a small dataset, significantly reducing training data and time while maintaining state-of-the-art performance.

Contribution

It demonstrates that effective face recognition quantization is possible with much smaller datasets and training times than traditionally required, challenging existing assumptions.

Findings

Achieved 96.15% accuracy on IJB-C dataset.

Used only 14,000 images for training, 440 times smaller than MS1M.

Established a new state-of-the-art in compressed face recognition models.

Abstract

In recent years, model quantization for face recognition has gained prominence. Traditionally, compressing models involved vast datasets like the 5.8 million-image MS1M dataset as well as extensive training times, raising the question of whether such data enormity is essential. This paper addresses this by introducing an efficiency-driven approach, fine-tuning the model with just up to 14,000 images, 440 times smaller than MS1M. We demonstrate that effective quantization is achievable with a smaller dataset, presenting a new paradigm. Moreover, we incorporate an evaluation-based metric loss and achieve an outstanding 96.15% accuracy on the IJB-C dataset, establishing a new state-of-the-art compressed model training for face recognition. The subsequent analysis delves into potential applications, emphasizing the transformative power of this approach. This paper advances model quantization by highlighting the efficiency and optimal results with small data and training time.