I am developing a model of object detection based on fast-rcnn architecture (transfer learning) in tensorflow object detection API. My problem is that created model happens to produce very good results when a searched object is close to camera (99% of frames) however it often fails to recognize objects that are placed in some bigger distance from the camera (50%). What I exactly mean is depicted on the figure:
My question is how can I improve the correct recognition on longer distances?
I may also add that my database that I used to train this model already contains a lot of labeled objects that are placed in the distance it has problems with.
[EDIT] I've just realized that my question could be simplified from "distanced objects" to "smaller objects", because from perspective of camera, objects that is further away is simply smaller