A script to automatically "block" people in images (like the Black Mirror episode White Christmas) using Mask R-CNN pretrained on the MS COCO dataset. No GPU required!

But you can block more than just people: up to 80 different types of objects can be blocked, including giraffes and busses!

This project relies on a handful of dependencies, use the following command to install your dependencies:

pip3 install -r requirements.txt

Note: Depending on your environment, you may need to use sudo. You may also want to use virtualenv.

Person Blocker is used from the command line:

python3 person_blocker.py -i images/img3.jpg -c '(128, 128, 128)' -o 'bus' 'truck'

• -i/--image: specifies the image file.
• -m/--model: path to the pretrained COCO model weights (default: current directory): if not specified, it will download them automatically to the current directory if not already present (note: the weights are 258 MB!)
• -c/--color: color of the mask, in either quote-wrapped hexidecimal or 3-element RGB tuple format. (default: white)
• -o/--object: list of types of objects to block (or object IDs of specific objects). You can see the allowable choices of objects to block in classes.py or by using the -names flag. (default: person)
• -l/--labeled: saves a labeled image annotated with detected objects and their object ID.
• -n/--names: prints the class options for objects, then exits.

The script outputs two images: a static (pun intended) image person_blocked.png and an animated image person_blocked.gif like the one at the beginning of this README.

python3 person_blocker.py -i images/kite.jpg

python3 person_blocker.py -i images/giraffe.jpg -c '#c0392b' -o 'zebra'

python3 person_blocker.py -i images/img3.jpg -c '(128, 128, 128)' -o 'bus' 'truck'

Blocking specific object(s) requires 2 steps: running in inference mode to get the object IDs for each object, and then blocking those object IDs.

python3 person_blocker.py -i images/img4.jpg -l

python3 person_blocker.py -i images/img4.jpg -o 1

The same requirements as Mask R-CNN:

• Python 3.4+
• TensorFlow 1.3+
• Keras 2.1.6
• Numpy, skimage, scipy, Pillow, cython, h5py

plus matplotlib and imageio

The credit for this code goes to @minimaxir. I've merely reproduced the results.