Fastai v2 - SSD/YOLO

fastai
1

Hi, Two questions regarding fastai v2 and SSD/YOLO.

  1. My understanding based on the previous course is that SSD/YOLO are manually implemented within fastai. Is this correct? I don’t think there is a pre-built model, but I do understand there is an NVIDIA model of SSD on the pytorch hub, but I’m assuming that they still had to build the SSD head on resnet like shown in the previous fastai course
  2. Since they switched to the new v2 API any suggestions on the starting point for implementation? I’m assuming you can just reimplement the old version but does the new one do this with callbacks and the like? The API is pretty fresh so doesn’t feel like there is much out there for it yet.
  3. I’m assuming the output from SSD/YOLO would match the mulicategory output for one portion, but you’d need a second for the bounding boxes too?

Thanks for the help!