v1.5.0

What's Changed

• Add replace with mosaic option by @Shazi199 in #42
• Fix crash when trying to keep audio but there's no audio stream by @borijang in #49
• Keep image metadata after blurring by @Frezs19 in #45
• Add basic black-box test using tox, automatically run with a new GitHub actions workflow

New Contributors

• @Shazi199 made their first contribution in #42
• @borijang made their first contribution in #49
• @Frezs19 made their first contribution in #45

Full Changelog: v1.4.0...v1.5.0

v1.4.0

• Accelerated onnxruntime execution providers like OpenVINO or DirectML are now
  automatically selected for inference if available. The presumably fastest
  provider is chosen by default. To override this choice you can use the new
  --execution-provider CLI argument (also fixes #40).
• Switch included ONNX model to optimized version with batch normalization ops merged into the adjacent convolution layers. This leads to slightly better performance while keeping the end results the same (#39).
• Update documentation according to these changes.
• Fix a bug that occurs when passing differently shaped inputs in the same deface call (#41).

Full Changelog: v1.3.0...v1.4.0

v1.3.0

What's Changed

• Add --keep-audio (-k) option to preserve / copy audio track of videos, as requested in #4. Default behavior is still that the audio track is discarded. (48b655e)
• Require imageio>=2.25 and imageio-ffmpeg>=0.4.6, which are necessary for the --keep-audio implementation.
• Modernize build system using pyproject.toml, use setuptools-scm instead of versioneer and remove a few setup files that are no longer needed now. (5ae6ded)
• Use a GitHub workflow for automatic package publishing on PyPI when releases are tagged. (79f4cb4)

Full Changelog: v1.2.0...v1.3.0

v1.2.0

What's Changed

• Support RGBA and grayscale input images (outputs are always RGB) in 852286e
• Fix ort explicit providers required by @Aziks0 in #22
• Fix inconsistent fps issues (fps reading vs fps writing) by @mysablehats in #28

New Contributors

• @Aziks0 made their first contribution in #22
• @mysablehats made their first contribution in #28

Full Changelog: v1.1.1...v1.2.0

v1.1.1

• Fix a regression of CLI input path parsing that was introduced with v1.1.0:
  • Invalid input paths were accidentally ignored instead of triggering error messages
  • The $ deface cam shortcut wasn't working properly

Full Changelog: v1.1.0...v1.1.1

v1.1.0

What's Changed

• Support directories as input, by @faaip in #10
• Add function to get anonymized image without CLI, by @faaip in #13
• Add option to replace faces with a custom image, by @medrimonia in #17
• Fix incompatibility with ONNX 1.9.0+ in c4b375b

New Contributors

• @faaip made their first contribution in #10
• @medrimonia made their first contribution in #17

Full Changelog: v1.0.0...v1.1.0

v1.0.0

This is the first major release of deface. The CLI has changed significantly since the last release, so please make sure to read the new documentation if you have used an older version before.

What's new

CLI

• Support multiple input paths, also via shell expansion. Directories should now be processed through shell expansion.
• Remove support for passing directories directly as inputs
• Remove support for file name filters.
• Don't show preview GUI by default.
• Add "cam" input shortcut for webcam demo with live preview.
• Don't start webcam demo by default if no input paths are given, because this had lead to strange behavior on systems without webcam hardware. A help message is shown instead and the user is prompted to give input paths.
• Change some option names for better clarity and simplicity. Please refer to https://github.com/ORB-HD/deface#cli-usage-and-options-summary for the new option names.
• Improve CLI argument help texts. Display default values for options.
• Enable complete control over the FFMPEG encoder configuration through the new --ffmpeg-config option. This controls how output videos are created.
• Improve reporting of input and output files that are currently being processed.

Docs

• Add detailed documentation to the readme.
• Add example images.

Other

• Modify detection score overlays to make them more readable, disable face enumeration.
• Lift imageio-ffmpeg version restriction.
• Support preview GUI for photo inputs.
• Close preview GUI when the Esc or Q key is pressed, communicate this in the window title.
• Add a pyinstaller spec file. A pyinstaller distribution of deface can now be created by running $ pyinstaller deface.spec inside of the cloned repository.

v0.4.2

• Update URLs to the new repo location https://github.com/ORB-HD/deface.
  Previous releases can be found at https://github.com/mdraw/deface/releases
• Fix sdist packages for PyPI