Different Defects, Different Effects: A Controlled Audit of Face Recognition Training Data
Authors: Ana Dias, João Ribeiro Pinto, Hugo Proença, João C. Neves
arXiv: 2610.28766 [cs.CV]
Submitted: 20 October 2026 · Accepted at: WACV 2027
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
The development of face recognition has been strongly supported by the availability of large-scale training datasets. Among them, MS-Celeb-1M has played an important role by providing millions of facial images from nearly 100K identities. However, the size and heterogeneity of such datasets increase the likelihood of annotation errors, redundant samples, and other forms of data inconsistency. This motivated the introduction of refined versions, such as MS1MV2 and MS1MV3, aimed at improving training-data quality.
Despite these cleaning efforts, the authors show that several types of training-data defects remain in MS1MV3. By proposing complementary data-auditing strategies, they identify five defect categories:
- Exact and near-duplicate images
- Split identities
- Crop errors
- Multiple-face ambiguity
- Invalid content
The study then assesses the effect of correcting these defects under a controlled face-recognition training protocol based on IResNet-50 and ArcFace.
Key Findings
The results show that targeted interventions can preserve—and in some cases improve—recognition performance while requiring fewer optimization steps to reach the performance of the original training set.
- Duplicate removal provides the clearest overall benefit.
- Split-identity merging and invalid-content removal also improve training efficiency.
- Overall, the effect of dataset cleaning depends strongly on the type of defect being addressed.
These findings provide further evidence that training-data quality remains an important and comparatively underexplored factor in large-scale face recognition.
Availability
The training manifests and code used in this work will be released upon acceptance.
Metadata
| Field | Details |
|---|---|
| Comments | Accepted at WACV 2027 |
| Subjects | Computer Vision and Pattern Recognition (cs.CV) |
| Cite as | arXiv:2610.28766 [cs.CV] |
| DOI | https://doi.org/10.48550/arXiv.2610.28766 |
2026 context: As face recognition systems scale toward ever-larger pretraining corpora and increasingly agentic, multimodal pipelines, dataset auditing is becoming a first-class research concern. This work aligns with a broader 2026 push toward reproducible "data-centric" evaluation, where the provenance and cleanliness of training data are treated as measurable variables rather than fixed assumptions.
via ArXiv CV