Locate and detect salmon species

[KamiCreed]

Most of the dataset compilation code and dataset training is in the ./training folder. The dataset training is mostly a jupyter notebook and the dataset compilation code uses Datumaro within python scripts.

• Compile Kitwanga and Bear Creek data to YOLO format
• Try skipping frames to improve generalization
• Train YOLOv8n
• Convert nano model into TensorRT for Jetson Nano #76
• Train larger YOLO model or better
• Add male/female tag from the salmon annotations to the classes #52
• Convert CVAT exported annotations from PSF to track IDs and YOLO format #55
• Place a count threshold for each class for the test set #60
• Compare trained models with juvenile annotations added or removed #72
• Experiment on range of percentages of negative samples for training #79
• Compare models trained on only one river dataset against general
• Train a model with YOLOv9 for comparison

[KamiCreed]

Any completed optimizations or experiments should probably go into a future salmon paper. Let me know and I can give access to the overleaf paper.