ADNet implemented by Tensorflow
I implement it almost same with original flows.
If fact, the authors's code is something wrong. it is not same with his paper. So i think it is nearly upper-bound performance util the author upload right codes
Environments : Intel® Core™ i7-7700 CPU @ 3.60GHz × 8, GeForce GTX 1070
Prec. (20px) : (Paper : 85.1) (this code : 79.38)
IOU(AUC) : (Paper : 0.635) (this code : 0.601)
FPS : (Paper : 15FPS) (this code : 19FPS)
Tensorflow
Scipy
OpenCV
$ python tracker.py