search

theAIGuysCode / yolov4-deepsort

Object tracking implemented with YOLOv4, DeepSort, and TensorFlow.
GNU General Public License v3.0
1.33k stars 750 forks source link

Fix the tensorflow prediction (using models.load instead serving) #77

Open dac-nh opened 3 years ago

acschristoph commented 2 years ago

hey, when I use this fix, then fps drops from 15 to 3 FPS it looks like no gpu is using, any idea ? great project thanks

glenbhermon commented 1 year ago

hey, when I use this fix, then fps drops from 15 to 3 FPS it looks like no gpu is using, any idea ? great project thanks

Same here!

dac-nh commented 1 year ago

Hi, you should check line 11 to choose which GPU(s) are suite best for you. By default, I choose the first one, you can replace it by using all GPUs that you have by this: tf.config.experimental.set_memory_growth(physical_devices, True)

glenbhermon commented 1 year ago

Hi, you should check line 11 to choose which GPU(s) are suite best for you. By default, I choose the first one, you can replace it by using all GPUs that you have by this: tf.config.experimental.set_memory_growth(physical_devices, True)

Gives an error: ValueError: Unrecognized device: [PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')] the physical_devices[0] works, but is terribly slow at around 2 FPS (I'm using the Jetson AGX Orin)