# Identity Overlap Between Face Recognition Train/Test Data: Causing Optimistic Bias in Accuracy Measurement

arXiv: https://arxiv.org/abs/2405.09403 · HTML: https://mlsanity.com/paper/2405.09403/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 22 of its 35 references that are in the map.
- First posted on arXiv 15 May 2024.

## Builds on (most cited first, 22 of 22)

- [CosFace: Large Margin Cosine Loss for Deep Face Recognition](https://mlsanity.com/paper/1801.09414/) (2018)
- [VGGFace2: A dataset for recognising faces across pose and age](https://mlsanity.com/paper/1710.08092/) (2017)
- [SphereFace: Deep Hypersphere Embedding for Face Recognition](https://mlsanity.com/paper/1704.08063/) (2017)
- [MS-Celeb-1M: A Dataset and Benchmark for Large-Scale Face Recognition](https://mlsanity.com/paper/1607.08221/) (2016)
- [Learning Face Representation from Scratch](https://mlsanity.com/paper/1411.7923/) (2014)
- [Circle Loss: A Unified Perspective of Pair Similarity Optimization](https://mlsanity.com/paper/2002.10857/) (2020)
- [AdaFace: Quality Adaptive Margin for Face Recognition](https://mlsanity.com/paper/2204.00964/) (2022)
- [MagFace: A Universal Representation for Face Recognition and Quality Assessment](https://mlsanity.com/paper/2103.06627/) (2021)
- [CurricularFace: Adaptive Curriculum Learning Loss for Deep Face Recognition](https://mlsanity.com/paper/2004.00288/) (2020)
- [Cross-Age LFW: A Database for Studying Cross-Age Face Recognition in Unconstrained Environments](https://mlsanity.com/paper/1708.08197/) (2017)
- [NormFace: L2 Hypersphere Embedding for Face Verification](https://mlsanity.com/paper/1704.06369/) (2017)
- [Partial FC: Training 10 Million Identities on a Single Machine](https://mlsanity.com/paper/2010.05222/) (2020)
- [Face Recognition: Too Bias, or Not Too Bias?](https://mlsanity.com/paper/2002.06483/) (2020)
- [Towards Transferable Adversarial Attack against Deep Face Recognition](https://mlsanity.com/paper/2004.05790/) (2020)
- [ArcFace: Additive Angular Margin Loss for Deep Face Recognition](https://mlsanity.com/paper/1801.07698/) (2018)
- [WebFace260M: A Benchmark for Million-Scale Deep Face Recognition](https://mlsanity.com/paper/2204.10149/) (2022)
- [GroupFace: Learning Latent Groups and Constructing Group-based Representations for Face Recognition](https://mlsanity.com/paper/2005.10497/) (2020)
- [TransFace++: Rethinking the Face Recognition Paradigm with a Focus on Accuracy, Efficiency, and Security](https://mlsanity.com/paper/2308.10133/) (2023)
- [What Should Be Balanced in a "Balanced" Face Recognition Dataset?](https://mlsanity.com/paper/2304.09818/) (2023)
- [Consistency and Accuracy of CelebA Attribute Values](https://mlsanity.com/paper/2210.07356/) (2022)
- [MLFW: A Database for Face Recognition on Masked Faces](https://mlsanity.com/paper/2109.05804/) (2021)
- [Double Trouble? Impact and Detection of Duplicates in Face Image Datasets](https://mlsanity.com/paper/2401.14088/) (2024)

## Built on by (most cited first, 3 of 3)

- [Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors](https://mlsanity.com/paper/2409.02979/) (2024)
- [A Rapid Test for Accuracy and Bias of Face Recognition Technology](https://mlsanity.com/paper/2502.14996/) (2025)
- [Revising RVL-CDIP: Quantifying Errors and Test-Train Overlap](https://mlsanity.com/paper/2606.31446/) (2026)

## Abstract (arXiv)

A fundamental tenet of pattern recognition is that overlap between training and testing sets causes an optimistic accuracy estimate. Deep CNNs for face recognition are trained for N-way classification of the identities in the training set. Accuracy is commonly estimated as average 10-fold classification accuracy on image pairs from test sets such as LFW, CALFW, CPLFW, CFP-FP and AgeDB-30. Because train and test sets have been independently assembled, images and identities in any given test set may also be present in any given training set. In particular, our experiments reveal a surprising degree of identity and image overlap between the LFW family of test sets and the MS1MV2 training set. Our experiments also reveal identity label noise in MS1MV2. We compare accuracy achieved with same-size MS1MV2 subsets that are identity-disjoint and not identity-disjoint with LFW, to reveal the size of the optimistic bias. Using more challenging test sets from the LFW family, we find that the size of the optimistic bias is larger for more challenging test sets. Our results highlight the lack of and the need for identity-disjoint train and test methodology in face recognition research.