This week I researched some edge detection algorithms that we can use to filter out our image, wrote out one in matlab and one short one in python, since I believe both are compatible with Django. The sobel edge detector returns a smooth but a more faded image while the Canny detector returns a sharp contrast image. I will test them out on test data very soon to compare them. Next week, I’ll go on to test out which feature engineering and/or dimensionality reduction strategy to use so that we can move onto the neural network stage. Overall, I would say I made less progress on the project than I wanted to as I had a lot of assignments due and had to TA for ~20 hours for this particular week but hopefully next week, I’ll have some more time available.
Team Status Update for 10/10/2020
This week we all started working on different components of the project, the web app, edge detection and looking at the datasets to start building the neural network. Work on the web app had some issues that set us back but the server is now accessible on our tablets which is progress. We’ve done some experimentation with the OpenCV api in Django but we have fully implemented it into the app yet. The app is a crucial part of our project so it is important to get right as it ties together our deep learning and computer vision work. We also still need to create a github repository for synchronization across the different modules we are working on.
The progress with the neural net and image processing are also coming along, at a pace where we will be able to test and debug them in sync. Currently, we still need to have a team meeting about the neural net architecture and design. The main points of contention or inquiry are about specific libraries, the high-level layering of the net, and input data formatting. We also want to look into physically testing the edge detection we are writing, so we are looking into methods or ideas for this. However, the edge detection is just to speed up our net’s learning speed so we aren’t stressing over creating test cases for it specifically.
Aaron’s Status Update for 10/10/2020
This week I spent my time working with Django trying to get a web app up and running for our ASL interpreter. My goal was to get a web app up and running with the OpenCV api and make it accessible by our kindle fire tablets. I was able to get the server up and running but ran into an issue where you can’t run a publicly accessible server easily from WSL without doing some complicated network bridging. I tried to install Ubuntu on my computer as a workaround but that didn’t work. However I did find a method using the Termux on android that allows us to run the Django webapp straight off the tablet itself. Overall I think I made some headway but the setback didn’t allow me to make as much progress on the app as I wished. SO I’d say were a little behind on the Open CV integration but still relatively in line with our entire project schedule. By next week I hope to have a working Open CV integration in the app with a video feed preview.
Malcolm’s Status Update for 10/10/20
This week I spent time on an initial edge detection schema and managing the public datasets that we will be leveraging. This edge detection is fairly basic; it essentially uses Canny edge detection. The big issue with our edge detection is that we have no formal way of testing the efficiency of the edge detection. We would need to observe changes in the neural net in order to better tune the edge detection, so our approach to it is best-effort. Otherwise, I have made progress in identifying and creating frameworks for standardizing the data we need. Since our neural net is learning from three different data sets, it’s important to standardize them as much as possible. My goal for the coming week is to have the data standardized and to start experimenting with initial architectures for the neural net.
Team Status Update for 10/03/2020
This week as we are still in the preliminary stages of our project we decided to take the week to familiarize ourselves with Machine Learning/Deep Learning before we attempt to start implement our own neural networks for ASL recognition. We decided to go through the lectures from 10-601 Introduction to Machine Learning and go through the lectures for Regularization, Neural Networks, Backpropagation, and Deep Learning. Next week we hope to begin looking at Open CV and see how we can get it running on our tablets as well getting the ball rolling by looking through our data sets we found for our ASL recognition model and deciding which one to tackle first. We are currently on track with our project schedule and how to start working on the open cv and data set aspects of the project. Some risks or issues we foresee in the project is us spending a lot of time learning the concept required to implement our project so we’re keeping a close eye on the amount of time we dedicate to studying concepts and we hope to be able to put what we learned into practice with a relatively short turnaround. So far no significant changes have been made to the project and we don’t have any issues currently.
Lectures page: (https://scs.hosted.panopto.com/Panopto/Pages/Sessions/List.aspx#folderID=%229044a1d8-bf2d-4593-b478-a9d100e8a09f%22)
Aaron’s Status Update for 10/03/2020
This week as we are still in the preliminary stages of our project we decided to take the week to familiarize ourselves with Machine Learning/Deep Learning before we attempt to start implement our own neural networks for ASL recognition. We decided to go through the lectures from 10-601 Introduction to Machine Learning and go through the lectures for Regularization, Neural Networks, Backpropagation, and Deep Learning. So I went through and watched the lectures to try and get at least a base level understanding of the concepts. Some general things I took away was that we should avoid overfitting by making sure that our model doesn’t capture noise in our training data instead of the underlying features of the data and that more data points than features helps model behave and regularization helps when you can’t collect extra data. After this I think I still have some studying to do on deep learning concepts but I understand how were going to go about training our model a little better now.
Lectures page: (https://scs.hosted.panopto.com/Panopto/Pages/Sessions/List.aspx#folderID=%229044a1d8-bf2d-4593-b478-a9d100e8a09f%22)
Malcolm’s Status Update for 10/3/20
This week was spent both researching frameworks for neural networks and with getting familiar with the OpenCV library for python. We will be using OpenCV for pre-processing of our video data, since directly feeding video into our neural network would be much less efficient than using processed data. Mainly, I experimented with an attempt at edge detection in order to represent visual data in a more streamlined way. Our next steps at this point are centered around implementing OpenCV in a Django web-app to start working towards running our project from the tablets.
Young’s Status Update for 10/03/2020
This week mostly consisted of our learning and deep learning primer phase where we familiarise ourselves with the content that we’ll need to understand to implement the ideal ML algorithm we can use. Over the week I reviewed the 10-601 Introduction to Machine Learning recorded lectures and relevant homeworks such as neural networks and logistic regression. Antonis shared a 10-701 footprint recognition assignment with us because of the similarity to our project so I searched the assignment for inspiration on techniques we can incorporate. I also researched popular edge detection and feature extraction algorithms for the preprocessing stage of the workflow. Next week, we’ll plan to start implementing these two techniques and write different types of feature detectors so we can compare their performance later on.