Computer Vision with Michel — Week 3

Hey! My name’s Michel Liao. I’m a computer science first-year at Princeton University. I hope to get a Ph.D. in computer vision, publishing meaningful research along the way. Join me in my CV journey!

Github: https://github.com/Michel-Liao

Personal Website: https://michelliao.com/

Course Progress

Finished the first week of Coursera’s Advanced Learning Algorithms.

Paper of the Week

“Gradient-Based Learning Applied to Document Recognition”

Finals season! Postponing reading this until winter break.

Videos/Lectures

I’ve looked at cross-correlation and convolution with this video and these slides.

Assignments

I didn’t work on the assignment this week. My energy was spent on Coursera.

Insights

• Filters manipulate pixels in an image. On a basic level, they can create a filter in the colloquial sense, like on Snapchat. For image processing, however, they can highlight certain features in an image like edges. (Very useful for CNNs, apparently!)
• Cross-correlation is a method of applying filters. Cross-correlation is very similar to convolution, but the image arrays are reversed. I don’t really understand cross-correlation and convolution yet. I’ll learn more and write about this next week! (This article seems to be informative, but I haven’t read it.)
• Having a basic understanding of neural networks isn’t too challenging. You can think of neurons as logistic regression algorithms and layers as logistic regression, too.
• Implementing a basic neural network in TensorFlow is pretty easy. Just call the Sequential() function to instantiate the neural network. Inside the Sequential() function, call Dense() to create your layers. TensorFlow implements all the hard stuff for you!

Questions

• Where does the intuition for cross-correlation come from?
• Why/how does a Gaussian filter create a blurring effect?
• Do people still use MLPs for state-of-the-art neural networks, or are CNNs the go-to?

Going Forward

I’m focusing on my finals these next two weeks. I’ll work on Coursera and the assignment. If I have extra time, I’ll read the assignment.