I spent some time this week with the presentation and preparing for what I am going to say. This included creating the demo video that involved both recordings of the website interface and its ML capabilities as well as the rpi image retrieval system. Likewise, I had to compile all the testing results and graph them to portray in the slideshow. This consisted of gathering data that I had previously and learning how to plot using Matlab.
In addition to the design presentation, I focused a lot of my work on getting ResNet-18 to work and also simplifying its classification algorithm to output one of four results: orange, banana, strawberry, and canned food. This implementation was done solely on the web application rather than tested locally due to time constraints. I did some preliminary testing, and it worked enough to put in the results section. However, I need to focus more on making the image sent from the rpi to be clear enough to classify. This includes implement background reduction as well as an emphasis on the physical design to incorporate sufficient lighting to illuminate the image properly.
Lastly, I worked with Grace in setting up a SQL database that ensures safe access and editing of data. This is necessary in ensuring that the rpi data can be sent and stored to later be accessed by the website. To do this, we also did a lot of work in cloud deployment using Nginx which had its various complications. In the end, we were able to successfully deploy it, and now we just need to do testing on retrieval and processing through the cloud.
Next week, I plan to work closely with Surya in the communication between the rpi and the website. Likewise, I will need to help out with the placement of lights in the box to properly illuminate the image to prevent reflections and graininess. Then, I will need to tweak the ML classification and text recognition algorithms based off the original accuracy results. Lastly, I will need to work with Grace to ensure proper MySQL database access to ensure that data is properly being forwarded and processed according to our specifications.