Code to count coral spawn (eggs), run on central computer.
-
install CVAT locally: https://opencv.github.io/cvat/docs/administration/basics/installation/#ubuntu-1804-x86_64amd64
-
run
make_cslics_venv.sh -
conda activate cslics -
navigate to coral_spawn_counter folder
-
pip install -e .installs as a python package locally so other modules can use this code -
Will also want to install machine-toolbox (0.5.4)
-
NOTE, currently using old code and code in these files does NOT work with updated machine-toolbox
-
git clone https://github.com/petercorke/machinevision-toolbox-python.git -
cd machinevision-toolbox-python -
pip install -e .
conda activate cslicsto activate virtual environment- Open cvat by going to
http://localhost:8080in Google Chrome (cvat only works in Google Chrome) - Create project/tasks by creating spawn annotation/label and uploading the relevant images
- Once uploaded to cvat, export the dataset as a .zip file in cvat annotation format (should have
annotations.xmlfile)
- Assisted annotation via Hough Transforms for circular objects in the image, run
python sphere_annotations.py, which should take existing annotations.xml file and append all circles as bounding boxes for each image - Save new .xml file as a zip and re-upload to cvat (overwriting previos annotations)
- Check/view/modify annotations in cvat, download when annotations are ready for training
Detector fail cases (blank images) /home/java/Java/data/cslics_failurecases - 275 images currently, more can be sourced
2023 Dec Alor Tank4 cslics01 and 2023 Dec Alor tank3 cslics06 in /home/java/Java/data
other cslic runs are also on the SSD card /media/java/cslics_ssd
All cslics runs are in Rstore smb://rstore.qut.edu.au/projects/sef/marine_robotics/dorian/rrap/cslics
Current best ultralytics model /home/java/Java/ultralytics/runs/detect/train - aten_alor_2000/weights/best.pt trained on 2000 images of aten and alor cslics