This model detects beetles and scale bars in images by drawing bounding boxes around the respective items. This model was developed to facilitate downstream applications during BeetlePalooza 2024.

Model Details

yolo_beetle_best.pt is the weights file for the YOLO model. The yolov8m checkpoint was fine-tuned over 100 epochs on 29 annotated images of beetles sourced from the (2018-NEON-beetles dataset)[https://huggingface.co/datasets/imageomics/2018-NEON-beetles]. Please checkout the repository on HF and cite information accordingly.

All 29 images were used as the training set.

Model Description

The model is responsible for taking an input image (RGB) and generating bounding boxes for all classes below that are found in the image. Data augmentations applied during training include shear (10.0), scale (0.5), translate (0.1), fliplr (0.2), and flipud(0.2). The model was trained for 100 epochs with a default image size of 640.

Segmentation Classes

[box class] corresponding category

[0] beetle
[1] scale_bar

Details

model.train(data=YAML, 
            epochs=100, 
            batch=4,
            device=DEVICE,
            optimizer='auto',
            verbose=True,
            val=True,
            shear=10.0,
            scale=0.5, 
            translate=0.1,
            fliplr = 0.2,
            flipud = 0.2
            )

Metrics (Training)

              Class     Images  Instances      Box(P          R      mAP50  mAP50-95)
               all         29        479      0.992      0.998      0.995      0.743
            beetle         29        450      0.991      0.997      0.995      0.714
         scale_bar         29         29      0.992          1      0.995      0.771

Developed by: Michelle Ramirez

How to Get Started with the Model

To view applications of how to load in the model file and predict masks on images, please refer to the 2018-NEON-beetles-processing github page

imageomics/yolo_beetle_detection

Model Card for yolo_beetle_detection