# Towards Flops-constrained Face Recognition

arXiv: https://arxiv.org/abs/1909.00632 · HTML: https://mlsanity.com/paper/1909.00632/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 2 Sep 2019.

## Built on by (most cited first, 9 of 9)

- [Face.evoLVe: A High-Performance Face Recognition Library](https://mlsanity.com/paper/2107.08621/) (2021)
- [Decoupled DETR: Spatially Disentangling Localization and Classification for Improved End-to-End Object Detection](https://mlsanity.com/paper/2310.15955/) (2023)
- [Prototype Memory for Large-scale Face Representation Learning](https://mlsanity.com/paper/2105.02103/) (2021)
- [Towards Robust Face Recognition with Comprehensive Search](https://mlsanity.com/paper/2208.13600/) (2022)
- [BioNetExplorer: Architecture-Space Exploration of Bio-Signal Processing Deep Neural Networks for Wearables](https://mlsanity.com/paper/2109.02909/) (2021)
- [Rethinking Robust Representation Learning Under Fine-grained Noisy Faces](https://mlsanity.com/paper/2208.04352/) (2022)
- [A Survey of Face Recognition](https://mlsanity.com/paper/2212.13038/) (2022)
- [Towards Large-scale Masked Face Recognition](https://mlsanity.com/paper/2310.16364/) (2023)
- [Quality-Aware Prototype Memory for Face Representation Learning](https://mlsanity.com/paper/2311.07734/) (2023)

## Abstract (arXiv)

Large scale face recognition is challenging especially when the computational budget is limited. Given a textitflops upper bound, the key is to find the optimal neural network architecture and optimization method. In this article, we briefly introduce the solutions of team 'trojans' for the ICCV19 - Lightweight Face Recognition Challenge~citelfr. The challenge requires each submission to be one single model with the computational budget no higher than 30 GFlops. We introduce a searched network architecture `Efficient PolyFace' based on the Flops constraint, a novel loss function `ArcNegFace', a novel frame aggregation method `QAN++', together with a bag of useful tricks in our implementation (augmentations, regular face, label smoothing, anchor finetuning, etc.). Our basic model, `Efficient PolyFace', takes 28.25 Gflops for the `deepglint-large' image-based track, and the `PolyFace+QAN++' solution takes 24.12 Gflops for the `iQiyi-large' video-based track. These two solutions achieve 94.198% @ 1e-8 and 72.981% @ 1e-4 in the two tracks respectively, which are the state-of-the-art results.