Double Trouble? Impact and Detection of Duplicates in Face Image Datasets
| Autor: | Schlett, Torsten, Rathgeb, Christian, Tapia, Juan, Busch, Christoph |
|---|---|
| Rok vydání: | 2024 |
| Předmět: | |
| Druh dokumentu: | Working Paper |
| DOI: | 10.5220/0012422500003654 |
| Popis: | Various face image datasets intended for facial biometrics research were created via web-scraping, i.e. the collection of images publicly available on the internet. This work presents an approach to detect both exactly and nearly identical face image duplicates, using file and image hashes. The approach is extended through the use of face image preprocessing. Additional steps based on face recognition and face image quality assessment models reduce false positives, and facilitate the deduplication of the face images both for intra- and inter-subject duplicate sets. The presented approach is applied to five datasets, namely LFW, TinyFace, Adience, CASIA-WebFace, and C-MS-Celeb (a cleaned MS-Celeb-1M variant). Duplicates are detected within every dataset, with hundreds to hundreds of thousands of duplicates for all except LFW. Face recognition and quality assessment experiments indicate a minor impact on the results through the duplicate removal. The final deduplication data is publicly available. Comment: Accepted at the 13th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2024) |
| Databáze: | arXiv |
| Externí odkaz: |
|