Discord

Checklist AICC 2025 Round 1 · Task 1

The Defected Nuts

Produce pixel-level anomaly maps for defective hazelnut images when only defect-free images are available for training.

  • Vision
  • Unsupervised anomaly segmentation

The task

A chocolate factory's hazelnut inspection misses subtle defects such as hairline cracks, small cuts, holes and faint print contamination. A new system must flag defects at pixel level.

The training set contains 431 defect-free RGB PNG hazelnut images at 1024 × 1024 pixels; these are the only training data. The test set contains 70 images with defects (cracks, cuts, holes, print contamination) covering 0.23%–28% of the image, split 50/50 into Test A (visible leaderboard) and Test B (final leaderboard).

For each test image the contestant produces a 1024 × 1024 anomaly score map with values 0 (normal) to 255 (anomalous).

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
data/train/1.png431.png (defect-free) and data/test/1.png70.png (defective), all 1024 × 1024 RGB.
You submit
submission.csv with columns id (index following sorted filenames) and data (Base85-encoded bytes of a uint8 1024 × 1024 map, values 0–255); non-conforming submissions score 0.
Scoring
AUPRO (area under the per-region-overlap curve) up to a false-positive rate of 0.3, normalised by 1/0.3; each connected defect region is evaluated independently; range [0, 1], higher is better. The baseline scores 0.7492 on Leaderboard B.
Rules
  • The only pretrained model allowed is a standard ResNet18 pretrained on ImageNet (optional); no other pretrained models or datasets.
  • Individual participation (maximum team size 1); at most 20 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 1 (November 2025), online on Kaggle; Kaggle competition window 15 Nov 2025 (about 18:50 UTC) – 16 Nov 2025 19:00 UTC.

Details

Year
2025, Online (Kaggle)
Round
Round 1 · Task 1
Language
English
License
MIT (Kaggle licence of all three competitions). Solutions repository: MIT., as stated by the source