vito - Vision Tools

Python utilities for common computer vision tasks. The goal of this package is to provide a lightweight python-only package for very basic image manipulation tasks, optical flow handling, pseudocoloring and camera geometry.

Examples

• Pseudocoloring:

  from vito import imutils
  from vito import imvis
  
  # Load a single-channel image (data.dtype will be numpy.uint8)
  peaks = imutils.imread('peaks.png', mode='L')
  # Colorize it
  colorized = imvis.pseudocolor(peaks, limits=None, color_map=colormaps.colormap_viridis_rgb)
  imvis.imshow(colorized)
  
  # Load 16-bit depth stored as PNG (data.dtype will be numpy.int32)
  depth = imutils.imread('depth.png')
  # Colorize it
  colorized = imvis.pseudocolor(depth, limits=None, color_map=colormaps.colormap_turbo_rgb)
  imvis.imshow(colorized)

  Exemplary visualizations: colorizing via the turbo rainbow colormap (left); same data reduced to 11 bins colorized using viridis (right). Input data is obtained from two translated and scaled Gaussian distributions.
• Optical flow:

  from vito import flowutils
  from vito import imvis
  
  # Load optical flow file
  flow = flowutils.floread('color_wheel.flo')
  # Colorize it
  colorized = flowutils.colorize_flow(flow)
  imvis.imshow(colorized)

  Exemplary visualization: Optical flow (standard color wheel visualization) and corresponding RGB frame for one frame of the MPI Sintel Flow dataset.
• Pixelation:

  from vito import imutils
  from vito import imvis
  
  img = imutils.imread('homer.png')
  rects = [(80, 50, 67, 84), (257, 50, 82, 75)]  # (Left, Top, Width, Height)
  anon = imutils.apply_on_bboxes(img, rects, imutils.pixelate)
  imvis.imshow(anon)

  Exemplary visualization: anonymization example using imutils.apply_on_bboxes() as shown above, with Gaussian blur kernel (imutils.gaussian_blur(), left) and pixelation (imutils.pixelate(), right), respectively.
• For more examples (or if you prefer having a simple GUI to change visualization/analyze your data), see also the iminspect package (which uses vito under the hood).

Dependencies

• numpy
• Pillow

Changelog

• 1.6.1
  • House keeping: Internal adaptations to fix deprecated (NumPy) and changed (PIL) library usage.
  • CI updates to PyPI's trusted publishing.
  • Code clean up (added type hints, docstrings, ...).
• 1.6.0
  • Remove deprecated functionality (workarounds for Python versions older than 3.6).
  • Add type hints and update documentation.
  • Add parameter order check to imsave (that's the one thing I alwys mess up when coding without IDE).
  • imsave now creates the output directory structure if it did not exist.
• 1.5.3
  • Mute initial notification about imshow backend.
  • Catch exception and adjust imshow backend if used with headless OpenCV.
• 1.5.2
  • Changed imshow to skip waiting for user input if wait_ms=0.
  • Added (RGB) colormap handles for convenience.
  • Drop support for deprecated Python version 3.5.
• 1.5.1
  • Changed handling of None inputs for image concatenation/stacking.
• 1.5.0
  • Extends the imutils submodule (rotation, concatenation & noop).
  • Fix deprecation warnings with newer NumPy versions.
  • Extended test suite.
• 1.4.1
  • Removes f-strings to fix compatibility for older python 3.5.
• 1.4.0
  • Changed imvis.overlay to use a more intuitive signature.
  • Aliases for some cam_projections functions.
  • Spell-checked all files via pyspelling.
• 1.3.4
  • Extended input handling for imutils (support single channel input to rgb2gray).
  • Aliases for some imutils functions.
  • Cleaning up tests, documentation, etc.
• 1.3.3
  • Prevent nan values caused by floating point precision issues in the cam_projections submodule.
  • Remove the (empty) tracking submodule (to be added in a future release).
  • Update the submodule list.
• 1.3.2
  • Support custom label maps in detection2d module.
  • Construct BoundingBoxes from relative representations.
• 1.3.1
  • Relative BoundingBox representation.
  • Support label lookup for Detection instances directly.
• 1.3.0
  • Common representations and utilities for 2D object detection via the detection2d module.
    • Detection class to encapsulate object detections.
    • BoundingBox class to work with axis-aligned bounding boxes.
• 1.2.3
  • Support sampling from colormaps.
  • Adjust tests to updated PIL version.
• 1.2.2
  • Use explicit copies in pseudocolor() to prevent immutable assignment destination errors.
• 1.2.1
  • Explicitly handle invalid (NaN and infinite) inputs to pseudocolor().
• 1.2.0
  • Add pixelation functionality for anonymization via imutils.
  • Add Gaussian blur to imutils.
• 1.1.5
  • Extend projection utils.
• 1.1.4
  • Explicitly handle None inputs to imutils.
• 1.1.3
  • Fix transparent borders when padding.
• 1.1.2
  • Add sanity checks to imutils which prevent interpreting optional PIL/cv2 parameters as custom parameters.
  • Add grayscale conversion to imutils.
• 1.1.1
  • Maximum alpha channel value derived from data type.
• 1.1.0
  • Added padding functionality.
• 1.0.4
  • Improved test coverage.
  • Fixed potential future bugs - explicit handling of wrong/unexpected user inputs.
• 1.0.3
  • Minor bug fix: handle invalid user inputs in imvis.
• 1.0.2
  • Additional tests and minor improvements (potential bug fixes, especially for edge case inputs).
  • Ensure default image I/O parametrization always returns/saves/loads color images as RGB (even if OpenCV is available/used on your system).
• 1.0.1
  • Fix colorizing boolean masks (where mask[:] = True or mask[:] = False).
• 1.0.0
  • Rename flow package to flowutils.
• 0.3.2
  • Rename colorization for optical flow.
• 0.3.1
  • Fix colormaps.by_name() for grayscale.
• 0.3.0
  • apply_on_bboxes() now supports optional kwargs to be passed on to the user-defined function handle.
  • Changed imread()'s default mode parameter to optional kwargs which are passed on to Pillow.
  • Raising error for non-existing files in imread()
  • Added colormaps.by_name() functionality.
  • Fixed bounding box clipping off-by-one issue.
  • Added imutils tests ensuring proper data types.
• 0.2.0
  • Optical flow (Middlebury .flo format) I/O and visualization.
  • Support saving images.
  • Visualization utils for tracking results.
• 0.1.1
  • Changed supported python versions for legacy tests.
• 0.1.0
  • First actually useful release.
  • Contains most of the functionality of pvt (a library I developed throughout my studies).
    • cam_projections - projective geometry, lens distortion/rectification (Plumb Bob model), etc.
    • colormaps - colormap definitions for visualization (jet, parula, magma, viridis, etc.)
    • imutils - image loading, conversion, RoI handling (e.g. apply functions on several patches of an image).
    • imvis - visualization helpers, e.g. pseudocoloring or overlaying images.
    • pyutils - common python functions (timing code, string manipulation, list sorting/search, etc.)
• 0.0.1
  • Initial public release.
  • Contains common python/language and camera projection utils.