Project Overview

Data collection, preprocessing is a tedious process, definitely it takes a lot of time to collect good quality data, preprocees them (especialy image data of specific usecase ). To tackle this problem, I architected the pipeline, led, and developed a cross-platform Flutter application designed to streamline the data collection pipeline. The app enables users to acquire data and feed it into a model in just two simple steps, optimizing efficiency and usability. In the end, users (in our case robots) get the trained model ready for edge deployment.

Workflow

Alt Text

Key Features

  • Cross-Platform Compatibility: Developed using Flutter, ensuring seamless performance on both iOS and Android.
  • Streamlined Data Collection: Users can collect, label, and process data efficiently.
  • Automated Data Pipeline: The app integrates with cloud-based services to automate data processing and model training.
  • Minimal User Interaction: Designed to simplify the process, requiring only two steps to send data to the model.
  • Scalable Architecture: Built to support various data types and models for future adaptability.
  • End-to-End Model Generation: The pipeline not only processes data but also delivers a trained model at the end.

Technologies Used

  • Software Architecture: PlantUML.
  • Flutter (Dart) for cross-platform development
  • Firebase for cloud storage and backend services
  • Paperspace for fetching and training the data to produce the vision model.
  • CI/CD Pipelines for automated testing and deployment

Note: App is not open source and publicly not available in Appstore (Android or IOS), its an internal proprietory tool, for optimizing the data collection process.