vicoslab / overlay_viz

A python class for visualizing per-pixel predictions in deep learning.

Double-panel overlay visualizer

A python class for visualizing per-pixel overlay in to left/right-panel view.

A tool/class for double-panel visualization (left/right) of multiple overlays displayed over an image. Features:

• switch between multiple overlay (multiple channels) in left and right panel independently
• move between multiple images in folder
• toggle overlay and change alpha-mixing
• optionally load and display points
• user configurable (define main image filename suffix and suffix for left/right overlay filename)

Installation and dependencies:

Install as pip package:

pip install git+https://github.com/vicoslab/overlay_viz.git

Dependencies:

• Python >= 3.6
• NumPy >= 1.15.0
• Python OpenCV >= 4.5

Usage:

• Executable as a viewer for pre-generated results (saved as .npy files)

  overlay_viz --cfg demo/config.json --dir <data_folder>

  usage: overlay_viz [-h] [--cfg CONFIG_JSON] [--dir DIR]
  
  Overlay visualization tool.
  
  optional arguments:
    -h, --help  show this help message and exit
    --cfg C     config filename/path
    --dir D     main folder with images (will ask for folder confirmation)
  

• Class-based usage for a versatile way of viewing the predictions during learning (aka live view) - see demo.py

Config:

Configuration is loaded from the provided config.json where you can define:

• which filename suffix is used for main image
• which filename suffix (one or more) will be used for left and right panel overlays
• title of left/right panel
• background colors, loaded center points size, default blending factor, etc.

Default configuration is:

{
    "dx": 0.1, 
    "alpha": 0.5,
    "window_name": "image",
    "center_size": 2,
    "center_color": [ 0, 0, 255],
    "panel_title": {
        "left": "GT",
        "right": "pred"
    },
    "file_patterns": {
        "image": "0.img.png",
        "left_pane": ["GT.npy"],
        "right_pane": ["pred.npy"],
        "centers": "centers.npy"
    }
}

Input formats:

Pre-generated data for visualization can be NumPy arrays (.npy) or images (jpeg/png).

Expected NumPy data should be in the following format:

• rgb image (h x w x 3)
• left/right overlay arrays (c x h x w)
• optional:
  • object centers (n x 2)

All data must have the same corresponding size (h x w). Data that is stacked in 2D-grid image will be split into multiple patches for display in different channels, each with the same size as the input image.

Hotkeys:

'w/s' - switch left overlay channel index
'up/down' - switch right overlay channel index
'left/right' - previous/next image
'space' - toggle pause (during live view)
'+/-' - adjust overlay opacity
't' - toggle right overlay
'g' - toggle object centers display
'o' - select directory
'q' - quit

Example data:

Download example data from https://unilj-my.sharepoint.com/:f:/g/personal/jmuhovic_fe1_uni-lj_si/Emsoh0lHX-RGpEAEXJaA7ZwBRvxHkHThyE7Hfir95bSr7A?e=sl0aWp and run:

unzip visualizer_data.zip

overlay_viz --dir=./visualizer_data