Deep Learning

This section covers a detailed description of how to train a deep learning model to accomplish tasks in the Vision System.

Content

• How to Collect a Dataset
  • Guidelines for collecting a dataset
• Annotate your Dataset
  • Annotating Classification Data
    • Installing tkteach
    • Using tkteach
  • Annotating Segmentation Data
    • Installing LabelMe
    • Using LabelMe
    • Annotation Guidelines
• Setup Jenkins and FTP
  • Account Setup
  • Connecting to FTP
  • Logging in to Jenkins
• Train and Export a Model with Jenkins
  • Requirements
  • Creating a New Project
  • Training a Model
    • Classification Models
    • Segmentation Models
  • Training from a Previous checkpoint
  • Exporting a Model
• What to Look for While Training a Model
  • What the validation loss looks like in Jenkins
• Known Issues in the Jenkins System
  • ‘Early stopping: No improvement after 5 evaluations’
  • Resuming after Jenkins Restart
  • ‘Mean requires at least one data point’
  • ‘ParamScheduler RuntimeError’
• Test Your Model in the Bin Picking System
  • Setting up your Workspace
• Common Issues in the Bin Picking System
  • Memory Consumption issues
  • Deep Learning Node not able to run a second time
  • Models trained with the bottom layer removed
• Deep Learning Examples
  • Classification Model
  • Instance Segmentation Model
  • Key-point Detection Model
  • Suction Model