GUI — single image data editor

[ludimation]

MDD for data annotation application

• select and load image(s) using GUI—includes issue setup — load images function updates #37
• manipulate metadata—training mode — input custom "flow-skeleton" information #7, training mode — blow away the training data for an image #6, training mode — scale, position, and rotation GUI #5
• save data
  • per image
    • User-selected bone positions
    • kinect-input-based training data
      • absolute joint positions
      • images from kinect camera (one per joint set saved)
  • compile individual image data into conglomerated data and model files
    • specify
      • set of images (list of directories?) to compile
      • target directory & filename for saved data
      • data type for each joint (abs xyz, relative xyz, angle axis, etc.)
      • weights for each joint
      • prediction model / data type to compile
      • tags (wish list)
    • compile & save
      • training model file
      • metadata—xml file? or would csv / tsvs be quicker?
        • list of labels with associated file names+paths of images, and any tags
        • prediction model type
• load data
  • training model
  • metadata—setup — hashmap (or DB Table) to associate filenames with image labels #16
  • change prediction engine to match new data (could be done when creating a "corps" which uses that data)—debug overlay — GUI / keypresses for switching prediction data #24

[ludimation]

Immediate todo:

• debug map/pointer code
• image editor GUI setup
  • user index radio (up to 8)
  • image name field
  • random, next, previous buttons
  • save button
  • auto-generate skeleton from training data
• capture new training data using image editor
• save out image settings to XML file