Summary

Udacity's Self-Driving Car Engineer Nanodegree students share their experiences and methodologies in tackling neural network and computer vision projects, detailing their experimentation with TensorFlow, backpropagation, and the use of GPUs.

Abstract

The Udacity Self-Driving Car Engineer Nanodegree Program challenges students to complete projects that involve experimentation with various aspects of neural networks and computer vision. Students document their processes in blog posts, with Andrew Wilkie illustrating his use of TensorFlow and a range of functions for a Traffic Sign Classifier. Param Aggarwal describes an iterative approach, experimenting with 54 neural network architectures to improve classification accuracy. Jonathan Mitchell simplifies the complex concept of backpropagation, emphasizing its role in neural network training. Arnaldo Gunzi reflects on the power of GPUs in network training, drawing parallels to significant upgrades in everyday experiences. Lastly, Esmat Nabil explains the probabilistic Hough transform and its application in detecting lines within a video, highlighting the importance of parameter tuning for effective line detection.

Opinions

Andrew Wilkie values the exploratory nature of his project, highlighting the use of TensorFlow and various libraries in building his Traffic Sign Classifier.
Param Aggarwal expresses frustration at not achieving high accuracy despite using advanced convolutional layers, and the relief found in taking a break before continuing his work.
Jonathan Mitchell believes that understanding backpropagation is crucial and once mastered, it simplifies the rest of the deep learning process.
Arnaldo Gunzi is impressed by the performance boost provided by GPUs, likening it to transformative experiences such as finding a shortcut or receiving a reward.
Esmat Nabil emphasizes the efficiency of the probabilistic Hough transform in line detection and the necessity of fine-tuning parameters to detect small straight lines within curves.

Udacity Students Experiment with Neural Networks and Computer Vision

The Udacity Self-Driving Car Engineer Nanodegree Program requires students to complete a number of projects, and each project requires some experimentation from students to figure out a solution that works.

Here are five posts by Udacity students, outlining how they used experimentation to complete their projects.

Self-Driving Car Engineer Diary — 4

Andrew Wilkie

Andrew has lots of images in this blog post, including a spreadsheet of all the different functions he used in building his Traffic Sign Classifier with TensorFlow!

I got to explore TensorFlow and various libraries (see table below), different convolutional neural network models, pre-processing images, manipulating n-dimensional arrays and learning how to display results.

Intricacies of Traffic Sign Classification with TensorFlow

Param Aggarwal

In this post, Param goes step-by-step through his iterative process of finding the right combination of pre-processing, augmentation, and network architecture for classifying traffic signs. 54 neural network architectures in all!

I went crazy by this point, nothing I would do would push me into the 90% range. I wanted to cry. A basic linearly connected model was giving me 85% and here I am using the latest hotness of convolution layers and not able to match.

I took a nap.

Backpropagation Explained

Jonathan Mitchell

Backpropagation is the most difficult and mind-bending concept to understand about deep neural networks. After backpropagation, everything else is a piece of cake. In this concise post, Jonathan takes a crack and summarizing backpropagation in a few paragraphs.

When we are training a neural network we need to figure out how to alter a parameter to minimize the cost/loss. The first step is to find out what effect that parameter has on the loss. Then find the total loss up to that parameters point and perform the gradient descent update equation to that parameter.

Teaching a car to drive itself

Arnaldo Gunzi

Arnaldo presents a number of lessons he learned while designing an end-to-end network for driving in the Behavioral Cloning Project. In particular, he came to appreciate the power of GPUSs.

Using GPU is magic. Is like to give a Coke to someone in the desert. Or to buy a new car — the feeling of ‘how I was using that crap old one’. Or to find a shortcut in the route to the office: you’ll never use the long route again. Or to find a secret code in a game that give superpowers…

Robust Extrapolation of Lines in Video Using Probabilistic Hough Transform

Esmat Nabil

Esmat presents a well-organized outline of his Finding Lane Lines Porject and the computer vision pipeline that he used. In particular, he has a nice explanation of the Hough transform, which is a tricky concept!

The probabilistic Hough line transform more efficient implementation of Hough transform. It gives as output the extremes of the detected lines (x0, y0, x1, y1). It is difficult to detect straight lines which are part of a curve because they are very very small. For detecting such lines it is important to properly set all the parameters of Hough transform. Two of most important parameters are: Hough votes and maximum distance between points which are to be joined to make a line. Both parameters are set at their minimum value.