How to Implement Class Filtering During Prediction in SAHI Similar to Ultralytics YOLO?

[shamspias]

Hello

I am transitioning from Ultralytics YOLO to using SAHI for some of my projects and am trying to replicate a specific functionality. In Ultralytics YOLO, I can specify which classes the model should detect directly using a classes parameter during the prediction call. This setup allows the model to only process and return detections for the specified classes, ignoring all others, which is quite efficient.

Example of Ultralytics YOLO usage:

results = model(image, size=640, classes=[0, 2, 3])  # Detects only certain classes

I've been trying to find a similar feature in SAHI where I can limit the model to detect only specific classes at the time of prediction, but haven't found a way to do this yet. Here are my questions:

1. Is there a way in SAHI to set the model to only detect specified classes before the detection process starts, similar to the classes parameter in YOLO?
2. If such a feature doesn't exist, could you point me to the part of the SAHI codebase where I might implement this functionality?
3. Are there any potential impacts on performance or accuracy I should consider when modifying SAHI to add this feature?

This feature is crucial for applications where processing speed and focusing on specific objects are necessary. Any guidance or pointers you could provide would be greatly appreciated!

Thanks in advance for the help.