Datasets:

Cha53c
/

1kb-face-aligned

license: mit
task_categories:
  - image-classification
tags:
  - face-recognition
  - biometrics
  - image-compression
pretty_name: 1 kB Face - aligned verification crops (IJCB 2026)

Aligned face crops for Face Recognition at One Kilobyte (IJCB 2026)

Aligned face crops used by the IJCB 2026 paper Face Recognition at One Kilobyte: Evaluating Image Compression Algorithms for Barcode-Constrained Biometric Verification. Derived from the MIT-licensed AI-Solutions-KK/face_recognition_dataset (105 identities, 17,534 images) by face detection with a proprietary detector followed by ArcFace-style five-landmark alignment. Both crop resolutions (112 and 224 pixels) live in a single split; the resolution column distinguishes them.

Each record: image (PNG crop), identity (105 unique labels), file_name (source-relative name defining the canonical evaluation ordering), resolution (112 or 224).

from datasets import load_dataset

ds = load_dataset("Cha53c/1kb-face-aligned", split="train")
ds_112 = ds.filter(lambda r: r == 112, input_columns="resolution")

Replication code: https://github.com/innovatrics/1kb_face_ijcb