Detection using one stage detectors
nicolasj92 opened this issue · comments
Hi,
Did you consider evaluating this framework using a one stage detector such as yolo or ssd? Would the prediction work with the bounding box refinement of the anchor boxes of these approaches?
Thanks and best regards
Nicolas
We did not consider such object detectors for this project. However, any kind of detector which takes bounding boxes as input and applies regression/classification to them should work. The input has to be somewhere in the pipeline. Like the proposal generator in Faster R-CNN. In theory it should be possible to apply something like RetinaNet with Tracktor as well.