My environment with versions is listed at jitnet-environment.yml.
However, the main packages needed are:
detectron2
pytorch
torchvision
cv2
PIL
To run on a video:
python demo.py --demo /path/to/video --adaptive --save_video
Currently, I'm still training the COCO pretrained model, however you are free to use your own pretrained smaller model as a substitute for JITNet.