MegaFace is a set of public face datasets published and maintained by the University of Washington's Computer Science and Engineering Laboratory. It is a benchmark of 1 million-scale facial recognition algorithm. The dataset contains 690,572 images with 1,027,060 images. This repository contains a simple tool to evaluate your models on Megaface Challenge 1 benchmark. All the codes are implemented in Python and Mxnet in this repo.
This repository has been tested under the following environment:
- Python 2.7
- Ubuntu 18.04
- Mxnet-cu90 (==1.3.0)
- OpenCV 2.4.9
-
Prepare the environment. OpenCV 2.4 is required by the official devkit, for convenience, you can download it from BaiduCloud or GoogleDrive and unzip to
/usr/local/lib/opencv2.4. -
Clone the repository and put your models to be evaluated in
models/. -
Download megaface testpack data from BaiduCloud and unzip it to
data/. -
Download megaface official devkit from BaiduCloud or GoogleDrive and unzip to
devkit/. -
Edit and call
run.shto evaluate your face recognition model performance. -
Edit and run
plot_result.pyif you want to visualize the final results.
MIT LICENSE
@inproceedings{kemelmacher2016megaface,
title={The megaface benchmark: 1 million faces for recognition at scale},
author={Kemelmacher-Shlizerman, Ira and Seitz, Steven M and Miller, Daniel and Brossard, Evan},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
pages={4873--4882},
year={2016}
}
The code is adapted based on an intial fork from the insightface repository.