Gain insights into image data processing and CNNs for image classification using TensorFlow. Delve into CNN architectures and their applications, requiring Python and TensorFlow knowledge.

irml.tar.gz

image-rec

If you want to dive into the technology behind computer vision and self driving cars, this is where to start.

In this course you'll learn how to process data from image files and create convolutional neural networks (CNNs) to classify different types of images. After completing this course, you will have a solid understanding of why CNNs work and which CNN architectures to use for different tasks.

The code for this course is built around the TensorFlow framework, one of the premier frameworks for industry machine learning, and the Python pandas library for data analysis. Knowledge of Python and TensorFlow are prerequisites. 

This course was created by AdaptiLab, a company specializing in evaluating, sourcing, and upskilling enterprise machine learning talent. It is built in collaboration with industry machine learning experts from Google, Microsoft, Amazon, and Apple.

Image Recognition with Machine Learning

# Chapter Goals:

<ul>
	<li>Learn strategies for decreasing the number of parameters in a model</li>
	<li>Understand how the fire module works and why it's effective</li>
	<li>Write your own fire module function</li>
</ul>


<h2>A. Decreasing parameters</h2>


<p>In order to make a smaller model, we need to decrease the number of weights per convolution layer. There are three ways to decrease the number of weights in a convolution layer:
	<ul>
		<li>Decrease the kernel size</li>
		<li>Decrease the number of filters used</li>
		<li>Decrease the number of input channels</li>
	</ul>
We don't necessarily want to reduce the number of filters we use, since using a variety of filters allows us to extract different hidden features from the input. However, there are ways to decrease the kernel size and number of input channels while still maintaining good model performance.</p>
</div>


<h2>B. Kernel size</h2>


<p>The size of a kernel represents the amount of spatial information it can capture. For example, a 1x1 kernel will only capture the channel information for individual pixels, while a 3x3 kernel will aggregate the information between adjacent pixels within each 3x3 square of the input data.</p>
<p>Although larger kernels can capture more information, it comes at the cost of additional parameters. A convolution layer that uses 3x3 kernels will use 9x as many parameters as a layer that uses 1x1 kernels. A good strategy for balancing performance and parameter count is to use a mix of larger and smaller size kernels.</p>
</div>


<h2>C. Intermediate layer</h2>

The way we decrease the number of input channels is by adding an intermediate convolution layer. Though this may seem counter-intuitive, since adding an extra layer introduces additional kernel weights, it can drastically decrease the number of parameters used in a layer.
Consider a convolution layer with 100 filters and 3x3 kernels. Given the input has 50 channels, we can use the equation from chapter 1 to calculate the number of parameters in the layer:
$$
P = 3 \times 3 \times 100 \times 50 + 100 = 45{,}100
$$
Now, consider the case where we first apply an intermediate convolution layer with 10 filters and 1x1 kernels. The intermediate output will have 10 channels. The number of parameters used in the intermediate layer, $\small P_I$, is
$$
P_I = 1 \times 1 \times 10 \times 50 + 10 = 510
$$
Then if we pass the intermediate output into our original convolution layer, the total number of parameters used becomes
$$
P = P_I + 3 \times 3 \times 100 \times 10 + 100 = 9{,}610
$$
By adding the intermediate layer we decrease the number of parameters by a factor of nearly 5.
</p>
</div>


Learn about the central component of SqueezeNet, the fire module.

What you'll learn from this course

Image Processing

CNN

SqueezeNet

ResNet

Fire Module

Chapter Goals:

A. Decreasing parameters