FishingROV/scallop_yolo26x_lr_1280_v2

FishingROV YOLO26x L/R 1280 v2 (WIP)

Work in progress: this public release is for transparency and community testing. Metrics and deployment packaging may change.

Function

5-class scallop detector trained as a teacher/pseudo-labeler for FishingROV.

Classes:

• scallop
• king
• queen
• dead
• recessed

Input format (Left/Right)

Training/eval data are generated from 1920x1080 frames split into left and right panels, each rendered as 1280-scale inputs for detection (L/R panel workflow).

Current metrics

From run artifact results.csv (best mAP50 row):

• best mAP50: 0.4109
• mAP50-95 at best mAP50: 0.2664
• best epoch: 102

From class evaluation artifact class_eval_best.json:

• precision: 0.5303
• recall: 0.4118
• mAP50: 0.4317
• mAP50-95: 0.2723

Attribution & License

This model is a derivative work based on the University of St Andrews King Scallop dataset.

• Original DOI: 10.5281/zenodo.10156830

In accordance with the original dataset terms, this derivative work is released under the Creative Commons Attribution 4.0 International (CC-BY 4.0) license.

Included artifacts

• best.pt
• results.csv
• class_eval_best.json
• public_FishingROV_model_card_example.md (pulled from public FishingROV repo for provenance/template continuity)