Wagtail has the ability to automatically detect faces and features inside your images and crop the images to those features.
Feature detection uses third-party tools to detect faces/features in an image when the image is uploaded. The detected features are stored internally as a focal point in the focal_point_{x, y, width, height} fields on the Image model. These fields are used by the fill image filter when an image is rendered in a template to crop the image.
Two third-party tools are known to work with Wagtail: One based on OpenCV for general feature detection and one based on Rustface for face detection.
Feature detection requires OpenCV which can be a bit tricky to install as it’s not currently pip-installable.
There is more than one way to install these components, but in each case you will need to test that both OpenCV itself and the Python interface have been correctly installed.
opencv-python
opencv-python is available on PyPI. It includes a Python interface to OpenCV, as well as the statically-built OpenCV binaries themselves.
To install:
pip install opencv-python
Depending on what else is installed on your system, this may be all that is required. On lighter-weight Linux systems, you may need to identify and install missing system libraries (for example, a slim version of Debian Stretch requires libsm6 libxrender1 libxext6 to be installed with apt).
A system-level package can take care of all of the required components. Check what is available for your operating system. For example, python-opencv is available for Debian; it installs OpenCV itself, and sets up Python bindings.
However, it may make incorrect assumptions about how you’re using Python (for example, which version you’re using) - test as described below.
Test the installation:
python3 >>> import cv2
An error such as:
ImportError: libSM.so.6: cannot open shared object file: No such file or directory
indicates that a required system library (in this case libsm6) has not been installed.
On the other hand,
ModuleNotFoundError: No module named 'cv2'
means that the Python components have not been set up correctly in your Python environment.
If you don’t get an import error, installation has probably been successful.
Rustface is Python library with prebuilt wheel files provided for Linux and macOS. Although implemented in Rust it is pip-installable:
pip install wheel pip install rustface
Rustface provides a plug-in that needs to be registered with Willow.
This should be done somewhere that gets run on application startup:
from willow.registry import registry import rustface.willow registry.register_plugin(rustface.willow)
For example, in an app’s AppConfig.ready.
The face detection algorithm produces a focal area that is tightly cropped to the face rather than the whole head.
For images with a single face this can be okay in some cases (thumbnails for example), it might be overly tight for “headshots”. Image renditions can encompass more of the head by reducing the crop percentage (-c<percentage>), at the end of the resize-rule, down to as low as 0%:
{% image page.photo fill-200x200-c0 %}
Once installed, you need to set the WAGTAILIMAGES_FEATURE_DETECTION_ENABLED setting to True to automatically detect faces/features whenever a new image is uploaded in to Wagtail or when an image without a focal point is saved (this is done via a pre-save signal handler):
# settings.py WAGTAILIMAGES_FEATURE_DETECTION_ENABLED = True
If you already have images in your Wagtail site and would like to run feature detection on them, or you want to apply feature detection selectively when the WAGTAILIMAGES_FEATURE_DETECTION_ENABLED is set to False you can run it manually using the get_suggested_focal_point() method on the Image model.
For example, you can manually run feature detection on all images by running the following code in the python shell:
from wagtail.images import get_image_model
Image = get_image_model()
for image in Image.objects.all():
if not image.has_focal_point():
image.set_focal_point(image.get_suggested_focal_point())
image.save()
© 2014-present Torchbox Ltd and individual contributors.
All rights are reserved.
Licensed under the BSD License.
https://docs.wagtail.org/en/stable/advanced_topics/images/feature_detection.html