AI-plugins for phenopype - still under early development.

Currently, plugins for three libraries are available:

ultralytics

• predict_yolo_det (object detection): https://docs.ultralytics.com/tasks/detect/ (fine tuned models only)
• predict_yolo_seg (segmentation): https://docs.ultralytics.com/tasks/segment/ (fine tuned models, and the SAM-family [SAM, SAM2, mobileSAM, FastSAM])

pytorch

• predict_torch_seg (Torchvision segmentation models: https://pytorch.org/vision/main/models.html)

keras

• predict_keras (Keras segmentation models https://keras.io/examples/vision/oxford_pets_image_segmentation/)

Prior to using the functions with phenopype you need to install the respective dependencies:

1. Install phenopype (see https://www.phenopype.org/docs/installation/phenopype/ for more details):

pip install phenopype

2. Install the plugins module:

pip install phenopype-plugins

3. Install the dependencies

If you have a GPU and the appropriate drivers install, make sure you install a fitting CUDA version first - e.g., v12.1:

mamba install -c nvidia cuda-toolkit==12.1

1. With GPU support:

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu121

2. Without GPU support:

pip install torch torchvision

1. With GPU support:

pip install keras-gpu

2. Without GPU support:

   pip install keras

1. Install Ultralytics BEFORE phenopype due to conflicting opencv-python (ultralytics) and opencv-contrib-python (phenopype) versions (see step 2 for alternatives):

pip install torch torchvision ## needed
pip install ultralytics==8.3.64

2. If you have already installed phenopype and can't or don't want to uninstall it, you can do the following:

pip install ultralytics

## force reinstall opencv-contrib-python
pip install opencv-contrib-python --force-reinstall