Generating Precision, Recall, and [email protected] Metrics for Each Category in Faster R-CNN Using Detectron2 Object Detection Models

[Shahriar-RL]

Hi everyone,
I'm currently working on my computer vision object detection project and facing a major challenge with evaluation metrics. I'm using the Detectron2 framework to train Faster R-CNN and RetinaNet models, but I'm struggling to compute precision, recall, and [email protected] for each individual class/category.

By default, FasterRCNN in Detectron2 provides overall evaluation metrics for the model. However, I need detailed metrics like precision, recall, [email protected] for each class/category. These metrics are available in YOLO by default, and I am looking to achieve the same with Detectron2.

Can anyone guide me on how to generate these metrics or point me in the right direction?

Thanks for reading!