W600k-r50.onnx |best| -

Authors: Zhu, Y., Zhao, H., Zheng, M., et al.

The "R50" (ResNet-50) variant is often considered the "sweet spot" for production environments, offering near-state-of-the-art accuracy with faster inference times than larger models like R100. deepinsight/insightface - 2D and 3D Face Analysis Project w600k-r50.onnx

: IResNet-50 (the "r50" in the name), a high-performance variant of the ResNet-50 architecture optimized for deep face recognition tasks. Authors: Zhu, Y

This file is an ONNX export of a ResNet-50 model trained on 600,000 identities ( w600k ). It is optimized for generating high-quality face embeddings. The model accepts a pre-processed face crop and outputs a dense vector suitable for cosine similarity matching. It is recommended for production environments requiring a balance between inference speed and recognition accuracy. w600k-r50.onnx