Tensorflow Introduction and 3D Object Recognition


Zhiang Chen

zxc251@case.edu

1. Useful Python Packages and Tools

  • numpy
    • The input and output of tensorflow network(computational graph) is numpy.ndarray
    • Integrating C/C++
    • Effecient n-dimensional array function (filter,reshaping)
    • Easy to learn
  • matplotlib.pyplot
  • Jupyter Notebook
    • Web editor for Python and other programming languages(Octave, R, ...)
    • Notes and Codes
    • Debugging in several cells
    • Easy to install (pip/pip3 install ipython) and learn

2. Tensorflow

Though there is no strict rule to build and train a network with Tensorflow, 6 steps are highly recommended. https://github.com/ZhiangChen/deep_learning/blob/master/27_objects/lenet5/lenet5.ipynb

3. 3D Object Recognition

ROS & Tensorflow

  • rospy is a Python2 package for ROS. The rospy client API enables Python programmers to quickly interface with ROS Topics, Services, Actions, and Parameters.
  • Tensorflow for Python2

ROS Package: auto_recognition2

auto_recognition2/evaluate_image2