Skip to content

Latest commit

 

History

History
22 lines (22 loc) · 3.31 KB

README.md

File metadata and controls

22 lines (22 loc) · 3.31 KB

Mini Project 3

The repo for the third mini project under EC601 - Product Design in ECE course at Boston University.

Mobile Machine Learning

Introduction

Mobile machine learning has come a long way in the last few years. These days, with the advent of powerful mobile hardware, it has become possible to develop mobile applications with powerful machine learning features like face recognition or text completion. There are two leading frameworks for this purpose, CoreML for Apple Devices, and MLKit for iOS and Android devices. Mobile ML presents unique challenges, like limited memory and power, and speed requirements. The algorithms need to be optimized for speed, as mobile use cases generally require instantaneous feedback, and the algorithms need to be power and compute efficient to be practical in these settings.

CoreML

CoreML
CoreML is Apple's proprietary framework for embedded machine learning inference inside Apps on Apple devices like the iPhone, Apple Watch etc. Models can be developed using either the proprietary CreateML format from the XCode IDE or converted to this format from supported frameworks like PyTorch or Tensorflow. It is more mature of the two frameworks, and was released earlier. Core ML supports a variety of machine learning models, including neural networks, tree ensembles, support vector machines, and generalized linear models. Core ML requires the Core ML model format (models with a .mlmodel file extension). Using Create ML and your own data, you can train custom models to perform tasks like recognizing images, extracting meaning from text, or finding relationships between numerical values. Models trained using Create ML are in the Core ML model format and are ready to use in your app. Apple also provides several popular, open source models that are already in the Core ML model format, like FCRN-DepthPrediction Depth Estimation, MNIST Drawing Classification, Resnet50 Image Classification, YOLOv3 Object Detection etc.
App Architecture

Unlike MLKit, CoreML models can only be trained in the cloud and not served from the cloud. Its main objective is to enable on device machine learning, utilizing more simple models that can be run on-device.

MLKit MLKit

MLKit is Google's proprietary framework for embedded machine learning inference on Android and iOS devices. It uses a Firebase backend for model training. Inference can be done either locally or via the cloud. It has superseded Tensorflow lite, which was designed for use in low power devices, such as mobile phones, and is more feature rich than it.

MLKit

Unlike CoreML, MLKit can also be used in conjunction with Firebase to serve ML Models via the cloud to millions of end users. This enables some powerful models to be run, with compute shifted to the cloud, provided the use case is not very latency sensitive, one such example being artwork generation.

MLKit

App

Our app is an android app running a tensorflow lite model implementation of the Inception v4 model. It generates bounding boxes and classification scores for objects it saw in the camera feed. Screenshot 1 Screenshot 2