Gauri’s Status Report for 04/25

We basically finished all the independent parts of our project this week!   I worked on a few things:

• Our classifier had a 95%+ accuracy on just identifying left vs right.  However, I realized earlier this week that it would classify everything it saw as left or right (even heads/walls/etc)
• I added a dataset of celebrity heads and trained the classifier to figure out head/left/right which it now does with a 93% accuracy.  (I did attempt to teach it walls/rooms as well but that was too complicated since there is too much variety in the stock images of these that I could find online)
• Shrutika and I integrated this working classifier with the Pi camera and played with the framerate to figure out how to handle frames as fast as possible and also get a reasonably accurate prediction by taking the mode of n predictions.  We decided 2fps was good.
• I added multiprocessing to make the classifier make predictions on frames faster by utilizing 2 cores of the Pi simultaneously.  This reduced the lag a little.  Still a bit slow because the classifier is slow to run (it is a neural net after all).
• Shrutika and I also worked on testing the mic setup before and after she built the baffles (all on zoom haha) and we also added multiprocessing to this so that the 4 mic channels’ inputs would be processed simultaneously on the 4 cores.  The performance of the mics dramatically improved after adding the baffles.
• We all demo-ed the three separate parts to Professor Sullivan and Jens on Wednesday.
• We will integrate (which should take < 1hr) the three pieces once Neeti is done perfecting the animation.  We are now working on the final presentation for Monday.

Overall we have almost wrapped up this project very well   Looking forward to finishing this up successfully!