Attention

This documentation is for an archived content. For the most up-to-date information, please visit the current FTC Docs.

FIRST Machine Learning Toolchain

Warning

Please be aware, TensorFlow has multiple “tasks” that it can perform - among these are “Object Detection”, “Image Classification”, “Speech Recognition”, “Segmentation”, “Natural Language Question Answering”, “Audio Classification”, “Optical Character Recognition”, and more. In FIRST Tech Challenge only the “Object Detection” task is supported - that allows for detecting multiple objects in an image along with bounding boxes to help identify where in the image the objects are found. Many online tools, such as Google Teachable Machines, use the “Image Classification” task - that allows for detecting a single object without a bounding box. These may seem similar, but they are not interchangeable. The FIRST Tech Challenge TensorFlow SDK ONLY supports the use of “TensorFlow Object Detection (TFOD)”. ftc-ml is a tool that is supported by FIRST Tech Challenge - using outside tools, even those designed for TFOD, are not supported by FIRST Tech Challenge; because of the “research” classification of TensorFlow breaking changes are inevitable and maintenance of projects in the TensorFlow community is abysmal. NO support will be provided for outside model trainers.

This tool, the FIRST Tech Challenge Machine Learning toolchain (FTC-ML), allows FIRST Tech Challenge teams to create custom TensorFlow models for use in the game challenge. Learn how to train TensorFlow to recognize your Custom Signal Sleeve images and more using this tool, and download models that you can use in your autonomous and driver-controlled Op Modes.

• Machine Learning In a Nutshell
• Logging in to FIRST Tech Challenge Machine Learning
  • Logging into the ftc-ml tool
  • Changing the active team login session
  • Logging out
  • Adding students to your team’s ftc-ml workspace
• Managing the ftc-ml tool workflow
  • Overview
  • Creating videos of objects to be recognized
  • Uploading videos to the ftc-ml tool
  • Adding labels to frames in a video
  • Producing Datasets
  • Training Models
  • Continuing Training on Models
  • Understanding Model Metrics
  • Canceling Training
  • Deleting a Model
  • Downloading Models
• Implementing in Robot Code
• Optimizing Videos for increased TensorFlow Model Performance
• FAQ
  • Why is TensorFlow called “TensorFlow”?
  • How many frames of our object is enough to ensure a good model?
  • How do I know if my model is trained well?
  • Why does my team get a limited amount of model training time?
  • Why can’t I seem to get a 100% object detection prediction?
  • I read somewhere about a parameter I can tweak…
  • Can object bounding boxes overlap?
  • What are the limitations imposed within the ftc-ml tool? (PER TEAM)

Volunteer Special Thanks

The FIRST Tech Challenge staff would like to extend a special thanks to the following volunteers for their hard work and dedication toward this project:

• Liz Looney, Google – FIRST Machine Learning Toolchain lead developer

• Mr. Phil Malone – Model designer and platform tester

• Uday Vidyadharan, Team 7350 – Platform tester and Contributor

• Jacob Burroughs – Platform configuration and SSO

• Richard Lester – Platform UI improvements