Checklist EUROAI (CEOAI) 2026 Contest Day 1 · Task 2
TextSnap
Read short uppercase alphanumeric text from images: predict its length, its letter/digit/space pattern and the full normalised string.
The task
Each test datapoint is an image containing uppercase letters, digits and possibly spaces, on one or two lines. The text must be read in natural order and normalised: letters uppercase, spaces and the line break written as '_', digits copied exactly.
Subtask 1 predicts the length of the normalised text, Subtask 2 the character-type sequence (L for letter, N for digit, _ for space), and Subtask 3 the full normalised text. Training character examples are 32×32 images with font size 10; test images are 256×128 with font sizes between 10 and 50.
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
- Training character images and hidden test images (256×128 RGB) via the platform; starter kit with a predict function stub.
- You submit
- Submitted through Nitro Cloud: a predict(img) function receiving an RGB NumPy array and returning (length, type sequence, normalised text).
- Scoring
- Subtask 1 (30 pts): exact-match accuracy; 3 pts below 0.11, linear 3–30 between 0.11 and 0.59, 30 pts at ≥ 0.59. Subtask 2 (20 pts): mean Levenshtein similarity; 2 pts below 0.51, linear 2–20 up to 0.85. Subtask 3 (50 pts): mean Levenshtein similarity; 0 below 0.08, linear 0–50 up to 0.70.
- Rules
- Official test set is hidden; outputs must not contain leading or trailing spaces.
- Submission limit 50, two final submissions (platform settings).
- Format
- Contest Day 1, 16 July 2026, 5 hours (10:30–15:30 local time; Nitro AI Judge 07:30–12:32 UTC), on-site at CREIC, Cluj-Napoca; three tasks, 100 points each.