PollEncOuNt

The purpose of this project is to count the number of alive and dead pollens in an image.

Contents

• Installation

• Outputs Demonstration

• Usage

  • GUI

  • Python API

Installation

1. Package installs

pip install PollEncOuNt

Outputs Demonstration

Raw	Predicted

blue squares detected are alive, green squares detected ard dead.

count_result.csv summarizes the results after running the prediction process.

Filename	Alive	Dead
pollen_predicted.jpg	96	20

Usage

GUI

Launching the GUI

Open your terminal (Mac/Linux) or command line (Windows).

Launch the GUI by typing:

% peon

The PEON Main Menu will appear.

Main Menu

From the Actions menu, you can:

1. Select Train to open the Train GUI.

2. Select Predict to open the Predict GUI.

3. Exit the application.

Predict GUI

A Pre-trained Model can directly be used for prediction.

1. Open the Predict GUI:

• From the main menu, go to Actions > Predict.

2. Configure Prediction Settings:

• Image Files: Click the Browse button and select one or more image files for prediction.
• Model File: Click the Browse button to load the trained model file (.pt, .pth, or .onnx).
• Project Save Directory: Click the Browse button to select the directory where the prediction results will be saved.
• Parameter Settings:
  • Check the Save Images box to save the images with predictions.
  • Check the Save CSV box to save the results in a count_result.csv file.
  • Set the Confidence threshold (0-1) to filter detection results. Higher values result in fewer but more confident detections.
  • Set the Int./Union (IoU) threshold (0-1) used for non-maximum suppression. Higher values allow more overlapping bounding boxes.
  • Set the Image Size for inference.
  • Set the Max Detections per image.
  • Add additional YOLO training parameters by clicking the Add Parameter button.

3. Start Prediction:

• Click the Start Prediction button to begin inference.
• Real-time log updates will appear in the LOGS section.

4. Reset Prediction:

• Click the Reset button to clear inputs and logs.

Train GUI

1. Open the Train GUI:

• From the main menu, go to Actions > Train.

2. Configure Training Settings:

• Data YAML File: Click the Browse button and select a valid YAML file for training data.
• Project Save Directory: Click the Browse button and choose the directory where training outputs (e.g., logs, model checkpoints) will be saved.
• Model Selection:
  • For Pre-trained Models: Click the Browse button to select a .pt, .pth, or .onnx model file.
  • For YOLO Models: Use the dropdown menu to choose one of the YOLO models (e.g., yolov8n.pt).
• Parameter Settings:
  • Set the number of Epochs for training.
  • Add additional YOLO training parameters by clicking the Add Parameter button.

3. Start Training:

• Click the Start Training button to begin.
• Monitor the real-time log updates in the LOGS section.

4. Reset Training:

• To reset the form, click the Reset button. This clears all input fields and logs.

Python API

Predict

A pre-trained model can directly be used for prediction(path: ./model/best.pt).

from peon import peon_predict

results = peon_predict(
    img_files=["IMG1", "IMG2", ...],
    model_path="MODEL_PATH",
    save_dir="SAVE_DIR",
    save_img=True,
    save_csv=True,
    conf=0.0,
    iou=0.5,
    imgsz=1280,
    max_det=500,
    **kwargs
)

img_files: list[str]

• List of image path to conduct prediction.

model_path: str:

• Path to the YOLO model.

save_dir: str

• Path to the directory to save the results. Required if save_img or save_csv is True.

save_img: bool (optional, defaults True)

• Whether to save the predicted images.

save_csv: bool (optional, defaults True)

• Whether to save the results to a CSV file.

conf: float (optional, defaults 0.0)

• Confidence threshold for filtering detections. Range 0-1. Higher values result in fewer but more confident detections.

iou: float (optional, defaults 0.5)

• Intersection over Union (IoU) threshold for non-maximum suppression. Range 0-1. Higher values allow more overlapping bounding boxes.

imgsz: int or tuple (defaults 1280)

• Image size for inference as integer or (height, width).

max_det: int (defaults 500)

• Maximum number of detections allowed per image.

**kwargs

• More YOLO prediction parameters

Train

from peon import peon_train

peon_train(
    data_path="DATA_PATH",
    save_dir="SAVE_DIR",
    model_path = "yolov8m.pt",
    epochs=100,
    **kwargs
)

data_path: str

• Path to the data .YAML file.

save_dir: str

• Path to the directory to save the trained model.

model_path: str (optional, defaults "yolov8m.pt")

• Path to the YOLO model.

epochs: int (optional, defaults 100)

• Number of epochs for training.

**kwargs

• More YOLO training parameters