shallowtoil/DROL

DROL-BAN

Opened this issue · 3 comments

xyl-507 commented

hello！
There is a problem when I introduced the classifier and template update strategy into SiamBAN.
The speed is slow, such as 4.5 FPS, and the GPU utilization is 100%. But DROL-RPN is 35 FPS tested on my workstation. I don't know the reason.
It would be great if you could give me some suggestions！
Thanks in advance!

xyl-507 commented

Hello, the author. Thanks for the outstanding contribution. I transplanted DROL to SiamBAN. After debugging, it is also successful, and the accuracy has increased. However, there is a problem that this part of code must be commented out：
# Train filter
if hard_negative:
self.filter_optimizer.run(cfg.TRACK.HARD_NEGATIVE_CG_ITER)
elif (self.frame_num-1) % cfg.TRACK.TRAIN_SKIPPING == 0:
self.filter_optimizer.run(cfg.TRACK.CG_ITER)

I don't know the reason, would you please give me some suggestions?
Thanks in advance!

TianXY-MouF commented

hello！ There is a problem when I introduced the classifier and template update strategy into SiamBAN. The speed is slow, such as 4.5 FPS, and the GPU utilization is 100%. But DROL-RPN is 35 FPS tested on my workstation. I don't know the reason. It would be great if you could give me some suggestions！ Thanks in advance!

Hi, I have met the same question, Have you solved it?
It would be great if you could give me some suggestions！
Thanks in advance!

xyl-507 commented

Sorry, I still haven't solved it.
I still need to comment out those two pieces of code
`
Train filter

    if hard_negative:

          self.filter_optimizer.run(cfg.TRACK.HARD_NEGATIVE_CG_ITER)

   elif (self.frame_num-1) % cfg.TRACK.TRAIN_SKIPPING == 0:

     self.filter_optimizer.run(cfg.TRACK.CG_ITER)`