// the find
minimaxir/person-blocker
Automatically "block" people in images (like Black Mirror) using a pretrained neural network.
A command-line tool that uses Mask R-CNN to detect and black out people (or any of 80 COCO object classes) in images, producing both a static PNG and an animated GIF output. It's a fun novelty script — the Black Mirror reference is the whole pitch. Aimed at anyone who wants to anonymize photos or just mess around with instance segmentation without writing model code.
Mask R-CNN handles per-pixel instance segmentation so the blocking is precise, not just a bounding box. You can target specific object instances by ID rather than blanketing every detected object of a class — useful when you want to block one person in a crowd but not another. Dockerfile is included so the dependency hell (TF 1.x, Keras 2.0.8) is at least containable. The two-pass workflow (label run to get IDs, then block run) is simple and sensible.
This repo is a time capsule from 2018 pinned to TensorFlow 1.3 and Keras 2.0.8 — both are EOL and incompatible with anything current. Getting it running on a modern Python environment without Docker is a genuine archaeology project. The 258 MB COCO weights download happens to the current directory with no configuration, which is sloppy. And there's been no meaningful activity since 2022, so don't expect fixes — if the Docker image breaks, you're on your own.