Checklist Benin selection 2026 Phase 1 · Task 1
Road Segment Detection
Classify aerial images according to whether they contain a segment of road.
The task
Governments map roads so that services such as healthcare and education can reach people. Some roads are built by governments, but others are made by people, for example near farms or mines; such roads are not always recorded, so officials must check and map them. Better road maps help governments find missing roads and, for Benin, support the national digital-infrastructure goals of the SNIAM 2023–2027 strategy.
The data were collected from imagery with a resolution of 25 cm covering urban, rural and agricultural landscapes; there are 1,200 images in total, split into train and test sets. Dry river beds, railway tracks and power lines can look like roads and must be classified as 'not road'.
For each test image the contestant predicts whether it contains a road segment (Target 1) or not (Target 0), submitting a probability.
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
Images.zip(all images, 408,621,730 bytes);Train.csv(image IDs with labels);Test.csv(image IDs of the test set);SampleSubmission.csv;StarterNB_RoadSegment.ipynb(starter notebook, 491,092 bytes). Downloads require a Zindi account.- You submit
- CSV with columns Image_ID and Target (e.g. ID_D9ONL553, 0.13); row order does not matter, but the Image_ID values must match
Test.csv. - Scoring
- Area under the ROC curve (AUC). The public leaderboard uses about 30% of the test set and the private leaderboard the remaining 70%. Zindi benchmark submission: 0.898022532 (public), 0.914108158 (private).
- Rules
- Open to Beninese pupils (collèges and lycées) under 20 on the first day of the international competition; access to the challenge required a secret code.
- Individual participation only.
- 15 submissions per day, 200 in total; 2 submissions selected for the private leaderboard.
- Only open-source languages, tools and packages; only the provided datasets; openly available pre-trained models are allowed; AutoML tools are not permitted.
- Probabilities must not be thresholded or rounded.
- The top 75 on the private leaderboard must submit reproducible code within 24 hours.
- Format
- Benin National AI Olympiad 2026, Phase 1: online Zindi hackathon, 17 Apr – 1 May 2026. Top participants advance to Phase 2 of the national preparation programme, a three-month training programme leading to the final phase of the national olympiad.