Deploying Python-based Deep Learning models on mobile devices : An app to detect and plot potholes

Vishal Gupta (~vishal11)


  • Deep Learning models can be trained and tested using Python but can be hard to port outside Python environments such as Android devices.
  • Our posters delves into one such application.
  • We trained a model to detect potholes and speed-breakers using Python and Keras but later deployed the model on an Android device to make inference in real time.
    • Gather and label training data for potholes, speed-breakers and accidents
    • Train model to predict the above using accelerometer and gyroscope readings
    • Save and convert model and weight files to TF-lite format
    • Build basic Android application to read phone’s accelerometer and gyroscope reading make predictions realtime
  • (Optional) Use a mapping API to plot and track potholes


  • Python, Keras, Sequential Models
  • Basic knowledge of Java and Android

Speaker Info:

  • Vishal Gupta, Machine Learning Engineer at Freshworks
  • Bharat Raj, Associate Engineer (Autonomous Driving), Siemens

Speaker Links:

Id: 1625
Section: Data Science, Machine Learning and AI
Type: Poster
Target Audience: Intermediate
Last Updated: