Using this approch on segmented images rather than bounding boxes. about object-detection-metrics HOT 1 CLOSED

Dear @Gasethata ,

The approach you want to use is not the purpose of this repository. Here I developed metrics for object detections for different classes using bounding boxes.

It seems to me that your evaluation method is pixel-wise instead of using bounding boxes. You should try to look for metrics that are applied to semantic segmentation approach, and for the most popular metrics used in your particular case. But if you still want to use bounding boxes to evaluate the detected classes' regions, you can adapt my code to obtain bounding boxes of the pixel regions. If you do it for ground truth and detected images, you can use my code to obtain mAP. But again, I don't know if this is suited to your case.

Regards.

from object-detection-metrics.