Checklist AICC 2026 Round 7 · Task 2
Oriented Ship
Detect ships near shorelines in aerial and satellite images with oriented (rotated) bounding boxes.
The task
A maritime analytics company needs to detect ships near shorelines in aerial and satellite imagery. Because ships appear at arbitrary angles, horizontal bounding boxes include too much background, so the system must predict oriented bounding boxes.
The images were split randomly into an 80% training set (903 images, with labels) and a 20% evaluation set (226 images, labels hidden). Each training image has a label file whose rows give cx, cy (box centre, normalised), w, h (normalised size) and theta (rotation in radians).
For each evaluation image the contestant predicts all ship boxes with a confidence score.
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/train (903 JPEG images), images/val (226 JPEG images for evaluation) and labels/train (.txt files with cx cy w h theta per box).
- You submit
- CSV with columns Id (image filename) and result: boxes formatted as 'conf cx cy w h theta', separated by commas; an empty string if no ship is detected.
- Scoring
- [email protected] with rotated IoU: a prediction is a true positive if R-IoU(B_pred, B_gt) >= 0.5; mAP is the mean of the per-class average precision (Kaggle metric name: SkewIoU).
- Rules
- Standard pretrained backbones (for example ImageNet ResNet or EfficientNet) are allowed.
- External maritime or satellite datasets are not allowed.
- Models pretrained or fine-tuned on oriented object detection datasets (for example DOTA) are not allowed.
- Individual participation (maximum team size 1); at most 30 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 7, online on Kaggle, 22 May 2026 20:02 UTC – 24 May 2026 20:00 UTC.