tamim662 / YOLO-Fish

A robust fish detection model for real-time underwater fish detection in any marine environments.

YOLO-Fish

A Robust Fish Detection Model to Detect Fish in Realistic Underwater Environment.

Accepted At Ecological Informatics [Paper]

Dataset

Download the annotated DeepFish dataset from here
To download OzFish dataset, visit the link

Working Procedure

This projects was implemented using darknet framework for detection model. To know the setup requirements, how to compile and train on different OS, how to use command line everything you will get by visiting AlexeyAB/darknet github repository.

Evaluation Model

1. Download the test dataset and unzip.
   test dataset of deepfish
   test dataset of ozfish
   test dataset of merge dataset
2. Download the cfg file of a chosen model.
   cfg of different models
3. Download the Model's weights.(google-drive mirror)
   models trained on deepfish
   models trained on ozfish
   models trained on merge dataset
4. Content of the file obj.data should be

classes=1
train=data/train.txt
valid=data/test.txt
names=data/obj.names
backup=backup/

5. obj.names files should contain just a 'Fish' word.
6. Keep wieghts files in a backup directory and cfg files in cfg directory.
7. Run the command(linux) to evaluate map for a YOLO-Fish models.

./darknet detector map data/obj.data cfg/yolo-fish-2.cfg backup/merge_yolo-fish-2.weights  

Run the command to test on video for real-time detection

./darknet detector demo data/obj.data cfg/yolo-fish-2.cfg backup/merge_yolo-fish-2.weights input.mp4 -dont_show -ext_output -out_filename output.avi

For using network video-camera mjpeg-stream with any Android smartphone, check here