eg4000
commented
5 years ago Hi,
In our original implementation we only used the soft scores. In the current implementation which has a stronger backbone and more reliable localization results, we found out that is it best to combine the two scores to get richer information. The rationale is that the two scores can be viewed as two different complementary methods to measure the localization quality: The hard score predicts how much the patch in the box looks like an object, whereas the soft score predicts how much the box overlaps with the underlying object,
Regards, Eran.
Nice job. I'm interested in your implementation and find that in this line you use a weighted sum of objectness score and iou score. It puzzles me as I think it's a little weird to add a score which denotes whether a bounding box contains an object with a score which denotes how large the bounding box occupy with ground truth. Do I misunderstand it? https://github.com/eg4000/SKU110K_CVPR19/blob/53feeaba724df62c93a71ea271d4d58b476e2851/object_detector_retinanet/keras_retinanet/utils/predict_iou.py#L57