jkjung-avt
speed on GPU
HI, thanks for awesom work, I tested in image by ssd_mobilenet_v2_egohands model in 1080ti GPU, but the speed is slow, about 700ms for each image averagely.
Are there some thing wrong with me, Could you please tell me the idea speed ? how to improve it ? thanks in advance.
When I tested my trained 'ssd_mobilenet_v2_egohands' on NVIDIA Jetson TX2, its inference speed was about 28.7ms per image. Reference: https://jkjung-avt.github.io/tf-trt-on-nano/
I guess your tensorflow is running in CPU mode. Please check and make sure you are using GPU version of tensorflow.
I use this tensorflow-gpu 1.12.0, but some prompt :
2019-07-04 11:59:59.571104: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1511] Adding visible gpu devices: 0
2019-07-04 11:59:59.571141: I tensorflow/core/common_runtime/gpu/gpu_device.cc:982] Device interconnect StreamExecutor with strength 1 edge matrix:
2019-07-04 11:59:59.571147: I tensorflow/core/common_runtime/gpu/gpu_device.cc:988] 0
2019-07-04 11:59:59.571153: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1001] 0: N
2019-07-04 11:59:59.571247: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 10421 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1080 Ti, pci bus id: 0000:03:00.0, compute capability: 6.1)
Is it the reason for speed slow?
The log looks OK to me. Sorry, I don't have an idea about why the 'ssd_mobilenet_v2_egohands' model runs slowly on your platform for now...