I’m working on a real-time object detection system using YOLOv8 and Python. My system is designed to detect industrial pieces as they pass by in a lot and ensure they belong to a specified lot type. Each piece has a label in the format ‘X123.4567.122.12’, where ‘X123.4567’ indicate the type and which groupe the piece belongs to, and in that groupe all pieces are almost the same with a small difference , and ‘122.12’ are unique identifiers for a specifique piece.

I’m encountering a challenge with distinguishing between very similar objects. Specifically, I need to differentiate between pieces with very similar characteristics where the only difference might be subtle variations. For example, consider two pieces that are almost identical, with only minor differences in their features. [as u can see here those are two difference pieces and the only difference is the two holes deplacement piece 1 ].

[piece 2:]

I added more images of similar pieces to the training dataset, but the model still struggles to differentiate the subtle variations.and Applied various augmentation techniques to improve generalization, such as rotation and scaling, but it hasn’t significantly improved accuracy in distinguishing these pieces.and nth really worked . i’m rly new to this so i have no idea what kind approach i should follow for similar pieces .