devbum On our nutty world

weblog

12:25:54 am: Going to start chasing my childhood dream of building a secret laboratory in the basement.

12:41:30 am: Took inventory -

  • Raspberry Pi x 2
  • Arduino x 1
  • Old Windows Laptop
  • USB Hard drive 500GB
  • T.V.
  • Monitor
  • Camera Module
  • Miscellaneous Sensors
  • Rooted Android Phone

01:12:48 am: Made a plan. Going to repurpose the old laptop to also serve as a file storage, maybe set it up as a local git server. Going to try my hand at dual booting Windows 7 and Linux Mint or ArchLinux. I’d rather just wipe Windows and be done with it but the ROM Flashing tool Odin only works on Windows machines so that’s a downer. Going to start that later. For now, to start off, I want to set up facial recognition with the camera module and Raspberry Pi.

01:24:13 am: With the TensorFlow 1.5 release, TensorFlow Lite is in preview. It’s less resource intensive than TensorFlow for Mobile but also features less ops. ‘InceptionV3’, ‘MobileNets’ and ‘On Device Smart Reply’ are the models guaranteed to work out of the box. Need to do some research into how well InceptionV3 and MobileNets works with facial recogniton.

01:36:00 am: Since the first task is just recognizing my face, there’s not much data available. To get enough training data I could write a script to continuously take picture for a period during which I move around a lot to try and get all the different perspectives. Might be a good idea to look at different ways of augmenting the dataset. Some of the popular methods include rotating and flipping each image and adding PCA noise (AlexNet).

01:44:37 am: Might make more sense to look at structural computer vision methods rather than statistical.