Discord

Checklist IOAI PH 2026 Team Selection · Task 2

Holy Week Maritime Reconnaissance

Detect swimmers, boats and debris in coastal drone images and flag unseen anomaly objects using generative data or vision-language models.

  • Vision
  • Object detection with out-of-distribution anomaly identification

The task

The scenario is a Philippine Coast Guard drone operation during Holy Week. Training images are annotated with classes 0 Swimmer, 1 Boat and 2 Debris. The test images may also contain two anomaly classes absent from training, 8 'Soggy Bunny' and 9 'Painted Egg', which contestants are expected to find with synthetic data from diffusion models or zero-shot vision-language models such as CLIP.

Detected anomalies are reported with a 1-pixel 'Ghost Box' at (0, 0, 1, 1).

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_images/, test_images/, train_labels.csv (image_id, annotations 'class_id x1 y1 x2 y2'), sample_submission.csv.
You submit
CSV with image_id, PredictionString ('class_id confidence x1 y1 x2 y2' …) and easter_egg_found.
Scoring
Custom blend: 50% mAP at IoU 0.5 on the maritime classes and 50% categorical accuracy on anomaly identification; leaderboard split 27% public / 73% private.
Rules
  • Scores are valid only if produced by a Kaggle notebook; external CSV uploads prohibited.
  • All notebooks must be shared; manual coordinate entry leads to disqualification.
Format
IOAI Philippines 2026 Team Selection, Task 2 (private Kaggle community competition).

Details

Year
2026, Online (private Kaggle community competitions)
Round
Team Selection · Task 2
Language
English
License
MIT (repository LICENSE), as stated by the source