Object Detector(s) implementation w/o compilation


(Naman) #1

I’m new here, please feel to make suggestions.

Looking for official object detector(s) implementation which works w/o compilation in latest stable Pytorch.
Few examples

  • MaskR-CNN (possibility to train without mask ground truth?)
  • Faster R-CNN
  • R-FCN
    OR single stage detectors
  • YoloV3
  • SSD

Would require to ‘train’ and ‘infer’ on new datasets & classes.
It seems the https://github.com/facebookresearch/maskrcnn-benchmark requires compilation (referred as installation) and works with Pytorch nightly.