fizyr / keras-retinanet

Keras implementation of RetinaNet object detection.
Apache License 2.0
4.38k stars 1.96k forks source link

problem with multiple objects per image #1363

Closed avivzadok172 closed 4 years ago

avivzadok172 commented 4 years ago

Hello, first thank you very much for this helpful repository. I am using this algorithm for my own data set of mammographs in order to detect tumors. In my case there are a lot of images with multiple objects. I know that it is instructed to write a distinct row for each label, like there are two different images with a single object. and so i did. but then i thought about it and it seems that this feature of only one object per image is really bad, because in a lot of cases my algorithm might detect a true tumor, but because it is not the one that he compares to (only one label per image) it would suffer a great loss. someone here knows how to bypass this problem, or explain to me if my argument is false. Thanks in advance.