when i try infering my fasterrcnn_resnet50_fpn i only get exactly 100 annotations per image when there are around 200-1000 which should be detected. The bbox-annotations are usually 16x16 pixels big and my image size is 2048x3072. I am not sure if the model itself is limited to output only 100 or if the task is unsuitable for that size of annotations but i have not read it anywhere. What could cause that? I tried changing a view parameters as the AnchorSize and RPN.
The default number of detections_per_image is 100.
Refer docs torchvision.models.detection.faster_rcnn — Torchvision main documentation
feeling a little embarrassed now. Must have overread it at least two times. Thanks a lot anyway!