Roughly speaking links included here go with different major topics
of the course.
Early OpenCV Examples
- Lecture 3 Examples.
This zip files contains five early tutorials as well as ex00movie
that illustrates how to advance through frames of an mpeg video.
Interactive Fourier Transform Sites
- Tomáš
Bořil Fourier Series 3D. This site lets you control the
construction of different functions by manipulating the phase and
magnitude of constiuent parts of the signal. The visualization takes
advanatage of a 3D view that is clever and allows more information
be shown in a single presentation.
- Dave
Watts Ejectamenta Fourier Site. Dave Watts has built an
excellent 2D fourier transform visualization tool that allows one to
move backward and forward between the spatial domain, e.i. a
greyscale image, and the Frequency domain. For testing I prefer the
following image
of the CSU Oval. Pay attention to the Short
Instructions and in particular the guideance on how to
construct a low pass and a high pass filter. Also, if you want to
test your skill, supress the 'noise' consisting of a sinusoidal
disturbance in this image
of the letter B.
Tensorflow
The web contains many helpful tutorials on tensorflow. I have only
begun to scratch the surface. That caveat offered, I found these
helpful.
Tensorflow, Layers and Estimators
The Layers modul elevates the level of abstraction for constructing
and running graphs in general and convolutional neural networks in
particular. The examples and relateed material here represent a
significant focus for our work on CNNs.
- tf.layers
Module API description. Here we find the definitive defintion
of the Layers API. Useful for details, but not a good substitute for
the tutorial - see below.
- Layers
MNIST Tutorial. There is a a lot going on in this example, and
it clearly is constructed to help educate us on how to best take
advantage of the Layers Module.
- There is source code on GitHub
and for convenience here is a local
copy.
Tensorflow and Tensorboard
Tensorboard is a powerful GUI tool that facilitates detailed
summaries and hence analysis of models. Here are links to a tutorial
using the MNIST dataset to demonstrate the use of summaires in
tensorflow to support analysis of outcomes in tensorboard.
Even though this tutorial and the previous one illustrating Layers
and Estimators solve the same problem of recognizing MNIST hand drawn
digits, the structures of the models are different. Notably, the
model illustrating summaries and tensorboard does not use any
convolutional layers. Also, this example does not use Tensorflow
Estimators.