Model Documentation.md

Micronekton Detection Documentation

Steps for deployment of FathomNet model

FathomNet/MBARI-midwater-supercategory-detector

Create and activate Conda environment

You only need to create your Conda environment once - proceed to the "Run Model" step if you have already done this

in a terminal window navigate into the folder you want to create a virtual environment in

$ module use /g/data/hh5/public/modules
$ module load conda/analysis3
$ python3 -m venv NAME_OF_ENVIRONMENT --system-site-packages 
$ source NAME_OF_ENVIRONMENT/bin/activate

install any missing libraries

$ (NAME_OF_ENVIRONMENT) $ pip install ultralytics

Clone yolov5 into the Conda environment

in a terminal window

$ git clone https://github.com/ultralytics/yolov5
$ cd yolov5/
$ pip install -r requirements.txt

Check that yolov5 has been cloned and all packages are working in Conda environment

$ cd (NAME_OF_ENVIRONMENT)
$ python
>>> import torch
>>> model = torch.hub.load("ultralytics/yolov5", "yolov5s")  

return to the terminal window

$ exit()

Prepare to run the model

download weights file

upload the imagery you want to run the model over and use the path when running the model script below

Run the model

in a terminal window

$ cd NAME_OF_ENVIRONMENT
$ source bin/activate
$ cd ../yolov5/
$ python detect.py --weights /path/to/best.pt --source /path/to/images-or-video --save-txt --save-csv --save-crop

Outputs

Outputs saved into ..yolov5/runs/detect/exp(n) where n is the run number

To move the outputs to a different location

$ cd runs/detect/
$ rsync -ravzP ./exp /path/to/DESTINATION

1. Original image with bounding box predictions

2. Cropped bounding box of predicted classes for each image

3. .txt file with all the bounding box information for each image

link to text file for this image

4. csv file with predictions and confidence levels for each image in the processed batch

link to csv file for processed images in this batch