I need to make a color sensor. The input is 6 wavelengths of color data (6 values for each measurement) and I want to compute the corresponding CIELAB value (consists of 3 output variables) from these 6 values. For training I can have it scan color swatches with known CIELAB value. We could make a csv table to feed the training. In days past I would have used a neural net but I was wondering if this could be down in tensorflow lite? Would also be a learning experience for bigger things. No experience with it yet but very proficient in arduino and python.
Thanks for any advice.
Hi Vijay & welcome to the TF Forum!
TensorFlow is a pretty extensive computation platform that’s designed for deep learning tasks, e.g. neural networks, so you can certainly implement your neural net in TF. TensorFlow Lite is a mobile/edge-optimised runtime for TensorFlow models, so you can run your TensorFlow neural network or other model on a mobile or IoT device.
You can train a model using TF Lite, but it’s typically faster to train your network on a workstation or cloud service with more compute power (using “normal” TensorFlow), and then use TF Lite to run your model on your Arduino. The guide explains how to convert a model to use with TF Lite.
If you are looking for a pre-existing model to start using without training, check out TF Hub for some solutions in common problem domains.