Diary/2019-5-26

RaspberryPi 3B+にTensorFlow

コマンド一発だとインストールできなかったので，細々と指定．

sudo apt install libatlas-base-dev
sudo pip3 install -U numpy
sudo pip3 install -U grpcio
sudo pip3 install -U absl-py
sudo pip3 install -U py-cpuinfo
sudo pip3 install -U psutil
sudo pip3 install -U portpicker
sudo pip3 install -U grpcio
sudo pip3 install -U six
sudo pip3 install -U mock
sudo pip3 install -U requests
sudo pip3 isntall -U gast
sudo pip3 install -U h5py
sudo pip3 install -U astor
sudo pip3 install -U termcolor
sudo pip3 install -U tensorflow_estimator
sudo pip3 install -U keras-applications
sudo pip3 install -U keras-preprocessing
sudo pip3 install -U tensorboard
sudo pip3 install -U tensorflow

ベンチマークあれこれ

Jetson TX2 CPUのみ

$ ./darknet detect cfg/yolov2-tiny.cfg yolov2-tiny.weights data/dog.jpg
mask_scale: Using default '1.000000'
Loading weights from yolov2-tiny.weights...Done!
data/dog.jpg: Predicted in 3.572668 seconds.
dog: 82%
car: 74%
bicycle: 59%

Jetson TX2 CPU+GPU

$ ./darknet detect cfg/yolov2-tiny.cfg yolov2-tiny.weights data/dog.jpg
mask_scale: Using default '1.000000'
Loading weights from yolov2-tiny.weights...Done!
data/dog.jpg: Predicted in 0.094432 seconds.
dog: 82%
car: 74%
bicycle: 59%

RaspberryPi3B+

$ ./darknet detect cfg/yolov2-tiny.cfg yolov2-tiny.weights data/dog.jpg
mask_scale: Using default '1.000000'
Loading weights from yolov2-tiny.weights...Done!
data/dog.jpg: Predicted in 26.837525 seconds.

RaspberryPi3B+ +NNPACK

$ ./darknet detect cfg/yolov2-tiny.cfg yolov2-tiny.weights data/dog.jpg
mask_scale: Using default '1.000000'
Loading weights from yolov2-tiny.weights...Done!
data/dog.jpg: Predicted in 0.978015 seconds.
dog: 82%
car: 74%
bicycle: 59%

Jetson Nano CPU

$ ./darknet detect cfg/yolov2-tiny.cfg yolov2-tiny.weights data/dog.jpg
mask_scale: Using default '1.000000'
Loading weights from yolov2-tiny.weights...Done!
data/dog.jpg: Predicted in 3.807130 seconds.
dog: 82%
car: 74%
bicycle: 59%

Jetson Nano CPU + GPU

$ ./darknet detect cfg/yolov2-tiny.cfg yolov2-tiny.weights data/dog.jpg
mask_scale: Using default '1.000000'
Loading weights from yolov2-tiny.weights...Done!
data/dog.jpg: Predicted in 0.159219 seconds.
dog: 82%
car: 74%
bicycle: 59%