This week I downloaded the final version of the trained neural network to use for our project. We ended up achieving an accuracy of 82% for the Cat vs. Non-cat inference. I breadboarded the final version of our solenoid circuit using the 12V DC Wall Adapter. The circuit was failing the first time because the transistor was blown, but after I replaced the transistor everything was working well. Finally, I worked with Philip to debug some issues with getting the machine vision code running on the Jetson. It turns out it was mostly just version incompatibility issues between the Jetson and Tensorflow.
I was in Boston for the end of the week (Fri – Sun) and so I wasn’t able to accomplish much else.
Next week, I will work with the team to finish integration and testing and get TensorRT running on the Jetson to improve performance. I will also begin preparing visuals (charts, etc.) for the final presentation and paper.