Checklist Cyprus AI Camp 2026 IOAI Selection Camp · Task 10
Document Layout Analysis
On DocLayNet page images, classify the document type, detect layout elements and produce pixel-level element masks.
The task
A legal-tech company needs to process scanned documents. Task A classifies each page into 6 document types; Task B detects layout elements (text, title, table, figure, list, other) with boxes and confidences; Task C produces a pixel-wise mask (0 background, 1–6 element classes). Any subset of tasks may be attempted.
The data are 2,500 real 800×1100 document images (2,000 train, 500 test) from IBM DocLayNet, with its document categories, bounding boxes (11 classes mapped to 6) and masks derived from the boxes.
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
train/test_images.npy, image id arrays,train_labels_a.npy(Task A),task_b/train.jsonin COCO format (Task B),train_masks.npy(Task C),class_info.json,metadata.npy,sample_submission.csv.- You submit
- One CSV with columns
image_id, task,doc_type,category_id,bbox_x,bbox_y,bbox_w,bbox_h, confidence,rle_mask(rows per task as specified). - Scoring
- Total = A + B + C (max 100): A = max(0, min(30, (accuracy − 0.5) × 60)); B = max(0, min(40, ([email protected] − 0.2) × 50)); C = max(0, min(30, (mIoU − 0.3) × 43)).
- Rules
- Pre-trained models and additional data are not allowed unless the task says otherwise.
- No outside help; LLM chats, ChatGPT, Copilot and similar tools are prohibited (except Colab inline autocompletion).
- Only the Kaggle website and Google Colab may be used (no local IDEs); online library documentation is allowed.
- One Kaggle account per participant; leaderboard name '[City] Name Surname'.
- Format
- Cyprus AI Camp 2026 (IOAI 2026 selection, in person, 31 Jan–8 Feb 2026): Segmentation, released 7 Feb 2026, 09:00–12:00 EET, on Kaggle (two tasks per 3-hour session).