V16 - Roboboat 2026

Better version of V15. Trained on more data.
we lowkey overtrained V15

Based on yolo11s

Available Formats:

PyTorch (.pt)
ONNX (.onnx)

Performance Metrics:

Precision: 96%
Recall: 94%
mAP50: 96%
mAP50-95: 78%

Class ID	Class Name
0	black_buoy
1	black_cross
2	black_target_boat
3	black_triangle
4	blue_buoy
5	green_buoy
6	green_light_buoy
7	green_pole_buoy
8	red_buoy
9	red_light_buoy
10	red_pole_buoy
11	yellow_buoy
12	yellow_target_boat

V15 - Roboboat 2026

Added 4 new classes for Roboboat 2026:

• red_light_buoy
• green_light_buoy
• yellow_target_boat
• black_target_boat

Based on yolov11s.

Available Formats:

• PyTorch (.pt)
• ONNX (.onnx)

Performance Metrics:

• Precision: 96.57%
• Recall: 94.81%
• mAP50: 96.81%
• mAP50-95: 80.53%

Dataset
Model on Roboflow

V14

V14 is now based on yolo11. V14 is trained on the same dataset as V13 was.

V13

V13 works pretty well, but is still weak when there are lighting artifacts on the camera lens.

V12

V12

V11

Adds X new classes

• blue_circle
• blue_cross
• green_cross
• green_triangle
• red_cross
• rubber_duck

These classes are all tested except for rubber_duck, and do not work well.

V10

THIS VERSION RENAMES CLASSES (to decrease confusion)

• green_ball -> green_buoy
• yellow_ball -> yellow_buoy
• blue_ball -> blue_buoy
• red_ball -> red_buoy
• black_ball -> black_buoy
• red_pole -> red_pole_buoy
• green_pole -> green_pole_buoy

Added 3 new classes

• red_racquet_ball
• blue_racquet_ball
• yellow_racquet_ball

Improve detections for overlapping buoys.
Buoys that are far away can sometimes be mistaken for racquet balls. This shouldn't be an issue due to how the tasks are setup

V9

V9 is trained on our entire dataset (as of this being released)

V8

Ball buoys are detected a lot better now!

V7

• improved dock detection.

Attached are pretrained model weights ready for use.