Tensorflow Object Detection from video source instead if the test image1.jpg image2.jpg with video.avi instead
Ubuntu 16.04
A tedious installation process..
First I was need to plug in a NVIDIA graphic card then install NVIDIA stuff.. CUDA 9.0 (Pitfall 1. not CUDA 9.2) then install cuDNN 7.0.5.15 (Pitfall 2. not the latest 7.1.14.18)
install OpenCV 3.4 with CMakeGUI take several houers
then install
Tensorflow
https://www.tensorflow.org/install/install_linux
Pitfall 3. replace command
$ sudo apt-get install cuda-command-line-tools
do instead
$ sudo apt-get install cuda-command-line-tools-9-0
Pitfall 4. replace command
(tensorflow) olle@olle-Aspire-M3800:~/tensorflow/tensorflow$ pip install --upgrade tensorflow-gpu
with
$(tensorflow) olle@olle-Aspire-M3800:~/tensorflow/tensorflow$ pip install --upgrade tensorflow-gpu==1.8.0.
when testing Tensorflow installation
$ sudo apt-get install python-opencv
I was follow the installation from the https://github.com/tensorflow/models/tree/master/research/object_detection
but modify the code a little for video.avi input instead of image1.jpg ...
starting the program
put a video.avi file in the working folder
python object_detection_tutorial.py