Discord

Checklist AICC 2025 Round 2 · Task 3

Face Matching

For each of 15 celebrity reference images, find all other images in a small collection that show the same person, without face-specific models or libraries.

  • Vision
  • Image retrieval / identity grouping
Read the official statement

The task

A collection of images shows 15 different celebrities, each appearing a variable number of times (at least 5 photos each). One reference image per celebrity is given.

For each reference image, the contestant must list every other image in the collection that shows the same celebrity; the reference image itself must not be included.

Face-specific models and libraries are prohibited, so the task must be solved with general-purpose pretrained models or OpenCV built-ins.

Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.

At a glance

You get
ref_img.csv (15 reference image IDs) and images/ (108 JPEG images named with 3-digit IDs, according to the Data page).
You submit
submission.csv with columns ref_img and photos (pipe-separated 3-digit image IDs), exactly 15 rows.
Scoring
F1 score per reference image (balance of precision and recall), averaged over the 15 reference images.
Rules
  • Solutions must run with less than 16 GB VRAM, and the entire notebook must finish in under 30 minutes.
  • No pretrained models designed for face recognition or face feature extraction (for example FaceNet, ArcFace, VGGFace, DeepFace); general-purpose pretrained models (for example CLIP, ResNet, ViT) are allowed.
  • No face-specific libraries (for example face_recognition, dlib, MediaPipe Face); anything included in OpenCV may be used without additional downloads.
  • No manual labelling or grouping of images.
  • Individual participation (maximum team size 1); at most 100 submissions per day.
  • AICC contest rules (stated on each Kaggle rules page, not enforceable): no use of LLMs for writing code or getting task ideas; no internet use other than official library documentation and the contest platform; no communication with anyone during the contest; clarifications only via the #clarification-requests channel on the AICC Discord server.
Format
AICC Round 2, online on Kaggle, 28 Dec 2025 09:00 UTC – 30 Dec 2025 09:00 UTC.

Notebook

Competition

Solutions

Details

Year
2025, Online (Kaggle)
Round
Round 2 · Task 3
Language
English
License
Varies by task: Essay Gap and Face Matching — CC BY-NC-SA 4.0; Audio Demixing — MIT (Kaggle competition licences). Solutions repository: MIT., as stated by the source