Skip to content

Latest commit

 

History

History
76 lines (62 loc) · 2.19 KB

README.md

File metadata and controls

76 lines (62 loc) · 2.19 KB

DeepLabCutLive

GUI to run DeepLabCut on live video feed

Install

Launch console:

conda create -n dlc_live
conda activate dlc_live
conda install python=3.7 tensorflow-GPU==1.13.1

Install dlc live GUI packages:

pip install https://github.com/DeepLabCut/DeepLabCut-live.git
pip install https://github.com/DeepLabCut/DeepLabCut-live-GUI.git

Run dlc live gui:

dlclivegui

Configure GUI

  • Config: scroll down menu: Create new config
  • Camera: scroll down menu: Add new camera or video from browser
    • Edit Camera Settings:
      • Select the camera by picking its serial number.
      • Rotate/Crop the desired portion of the image
      • display_resize: to use less resources while displaying
    • Init Camera

  • Processor settings:

    • Chose your processor directory using Processor Dir
    • Pick the processor class your wrote (see appendix 1
    • Edit or Set Proc

  • DeepLabCut:

    • From scroll down menu: Add DLC or chose your model, these settings can be modified using Edit DLC Settings
      • Specify Name, browse to extracted DLC model folder
      • Specify resizing similar to display_resize to make the inference quicker.
    • Init DLC
    • Display DLC Keypoints, keypoints can be edited but need to be 'undisplayed' first to have an effect.

Record Session

  • Record:
    • Set Up Session -> Ready -> On -> Off -> Save Video

Appendices

  1. Processor: Use the processor folder as a template to create your own processor The default processor should contain:
  • init
  • process: takes in a pose, performs operations, and returns a pose
  • save: saves any internal data generated by the processor (such as timestamps for commands to external hardware)
class Processor(object):

    def __init__(self, **kwargs):
        pass

    def process(self, pose, **kwargs):
        return pose

    def save(self, file=''):
        return 0