If modality is CAMERA ONLY, model cann't perceive objects in 360 view, especially the back behind the car. I wonder if 360 annotations are used to evaluate model trained by CAMERA ONLY or, do you only use annotations that are in the field of view of cameras?
yes, 360 annotations are used. those annotations that are not visible in cameras are considered FNs which is true for all CAMERA_ALL models. So the metrics should still be valid.
If modality is CAMERA ONLY, model cann't perceive objects in 360 view, especially the back behind the car. I wonder if 360 annotations are used to evaluate model trained by CAMERA ONLY or, do you only use annotations that are in the field of view of cameras?