Skip to content

Latest commit

 

History

History
73 lines (49 loc) · 3.38 KB

Readme_Select.md

File metadata and controls

73 lines (49 loc) · 3.38 KB

Selection Tool

The selection tool is used for

  • In detail inspection of predictions the network made
  • Selecting erronous predictions for further labeling

teaser

Overview

  • On its left hand side of the tool one time step of the video data is shown

  • Images recorded simultaneous by the cameras are shown in a tiled view

  • On the right hand side of the tool various controls are offered:

    • 'This frame'

      • Gives control over which keypoints are shown by selecting them, by default all keypoints are shown
      • 'Delete all': Discards all predicted keypoints, i.e. the annotator should label all of them later on
      • 'Recover': Can Undo delete all, by loading the predictions

    • 'All frames'

      • Selects the time step shown
      • Allows sorting by id (i.e. in the order they were recorded) or by score (i.e. how confident the networks prediction was)

    • 'Selected frames'

      • Selected frames are shown in the initially empty list
      • Provides automatic selection tools ('Select uniform' or 'Select by score')

    • 'Output'

      • Once 'write frames' is clicked the selected frames are grabbed from video input and stored to disk
      • In the folder the selected video was located a new directory is created that contains the selected frames
      • The tool chooses the first folder of the format 'labeled_set%d' that does not exist yet for storage (i.e. labeled_set0, labeles_set1, ...)

    • 'Score histogram':

      • Shows the distribution of prediction scores over the sequence shown, ranging from not confident (0.0) to very confidente (1.0)
      • The red vertical line indicates where the current frame is located

Starting the tool

For starting the selection tool a model configuration file {MODEL_CFG} must be provided (so the tool knows which keypoint are defined and how to show them) as well as an indication towards the data that should be shown {DATA_PATH}

python select.py {MODEL_CFG} {DATA_PATH}

For example the largest part of the tutorials uses the rat model so

{MODEL_CFG}=config/model_rat.cfg.json

If predictions are already made for a video sequence, the corresponding prediction file can be specified by {DATA_PATH}, f.e. it could look something like

{DATA_PATH}=data/pred_run000__01.json

If there aren't any predictions for the videos available yet, f.e. because it is the initial run of a new task, then {DATA_PATH} is set as path to one of the videos constituting the sequence

{DATA_PATH}=data/pred_run000_cam1.avi

If you need to adapt the appearance of the Selection tool towards your screen setting please see 'frame_size' in config/viewer.cfg.json

If you need to deviate from the assumed default layout on how calibration files are located (M.json in the path of {DATA_PATH}), you can call select.py using the --calib_file argument, which specifies the path to the calibration file wrt the path of {DATA_PATH}.

Using the tool

  • 'Mouse wheel': Zooming in and out of the left hand side frames
  • 'Left mouse button': Translating the frames shown left, right, up or down. (Usually, used when zoomed in)
  • 'Right mouse button': Delete a prediction. It is deleted from all frames with one click.
  • 'Arrow right', ->: Next frame
  • 'Arrow left', <-: Previous frame
  • 'S': Select current frame
  • 'W': Write selected frames to disk