This week has been a roller coaster for our prototype. We spent the first half of the week preparing for the presentation. While preparing for the presentation, the ADC failed. We had to replace the ADC on the PCB with the evaluation board, but we could not retrieve photos in time for the presentation. In the second half of the week, I worked with Ethan to integrate the hardware in a new enclosure. As we turned everything on on Friday night, both the ADC and camera were not working. We spent all of Saturday debugging the ADC and camera and eventually fixed them by swapping modules and reconfiguring the ADC’s configuration register.
Sid and I decided we needed to take more training photos since the results varied under different background lighting. After looking at the trigger, I also decided to report accuracy in two parts: classification accuracy and trigger accuracy. Classification accuracy is the accuracy on the test set. The trigger accuracy is the percentage of captures that are perfectly aligned with the image. The effective accuracy should be the product of the two. We will manually test the system on a unopened card deck to obtain an effective accuracy metric.
Overall, I am on track to finish testing my subsystems.