5,179,510 aligned face images of 93,431 individuals form this dataset introduced for the Lightweight Face Recognition Challenge at ICCV 2019. All images were processed using RetinaFace for landmark detection and resized to 112x112 pixels. The dataset is hosted by gaunernst on Hugging Face and was last updated in October 2024.
Use Cases
- Training face recognition models based on the large number of identities and aligned images.
- Benchmarking model performance for the Lightweight Face Recognition Challenge based on the dataset's origin.
- Developing facial analysis algorithms based on the standardized 112x112 image size and alignment.
- Studying facial representation learning based on the dataset's scale of over 5 million images.
Strengths
- Contains 5,179,510 images, providing a large-scale resource.
- Includes 93,431 distinct identities, offering a wide variety of subjects.
- Images are uniformly processed: aligned by RetinaFace and resized to 112x112.
- Compressed version reduces storage requirements while preserving original image data.
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download.
- Column-level documentation is absent; field semantics must be inferred after download.
- Data may reflect bias inherent to the original MS-Celeb-1M collection process.
Provenance
- Source
- MS-Celeb-1M dataset, introduced for ICCV 2019 Lightweight Face Recognition Challenge.
- Collection Method
- Images aligned based on facial landmarks predicted by RetinaFace.
- Freshness
- Last updated 2024-10-25 05:07:02; freshness should be verified.