search

danielgordon10 / re3-tensorflow

Re3 in TensorFlow
GNU General Public License v3.0
88 stars 36 forks source link

Increase fps with less accuracy #2

Closed mirkow closed 4 years ago

mirkow commented 4 years ago

First, great tracker!

I run it on a raspberry pi 4, but there I only get 2.5 FPS. Is there a way to easily increase the FPS but maybe with less accuracy as a drawback?

Thanks!

mirkow commented 4 years ago

Found now https://github.com/danielgordon10/re3-pytorch, which is supposed to have a smaller model..I'll try it and update with results

mirkow commented 4 years ago

Update: The small model in the pytorch variant (only 1.2 fps) is even slower than the big model in the tensorflow variant on the raspberry pi 4.

Rajasekhar06 commented 4 years ago

How did you ran the tracker only demo_batch is working for me

mirkow commented 4 years ago

I could just execute it. But this issue is probably not the place to discuss that

danielgordon10 commented 4 years ago

I don't think there's many deep learning models you can run on a Raspberry Pi. You could maybe try a MobileNet/Shufflenet backbone. I don't see why they wouldn't work. But adding these is not something I plan on doing.