Pin Insulator Instance Segmentation

Train

• CLIPD dataset is an image detection dataset for insulators.
• Preprocessed dataset into one insulator class and split into train/valid splits with a 80/20 ratio.
• Processed bounding box information according to YOLO format.
• Bounding boxes were input to mobile SAM model to autogenerate masks
• Autogenerated masks were used to train YOLOv8 segmentation model with pretrained weights resulting in Final trained weights.

Test

• Uncomment and run the commented cell in test notebook (change the path to the image you want to test)

Results

• precision(B): 0.9519
• recall(B): 0.9644
• mAP50(B): 0.9607
• mAP50-95(B): 0.8044
• precision(M): 0.9519
• recall(M): 0.9644
• mAP50(M): 0.9641
• mAP50-95(M): 0.8358