Please Come Back Karen, They're My Kids Too 😳🙏
When you have a falling out with someone close, cherished pictures can suddenly become unwanted reminders of former friends. Not anymore! Levying machine vision and image processing techniques, we can completely cut them out of your life, as if they were never there to begin with. With our advanced inpainting algorithms, we can remove the ugly while keeping the rest intact.
Avoid bittersweet memories while utilizing the cutting edge in machine vision and image processing
Repress, Revise, Redact!
How It Works 🤔💭
We have 2 pipelines for editing photos, one using the Google Cloud Vision API to compute bounding boxes for all humans in the target photo, and one that uses a Region-Based Convolutional Neural Network (R-CNN) trained on the COCO dataset to perform instance segmentation on all people in the image. The Google Cloud Vision-based method is used to reduce compute time (so Sean can run it on his laptop) but the R-CNN based method is more precise. Examples using both are in the images section.
From there, we compute bounding boxes for the faces of all detected people, and match with a known face of the person to be cropped out (using python face-recognition). We match the face bounding box with the closest human instance (using the 2-norm), then use OpenCV's TELEA inpainting algorithm to redact the pesky ex from the photo!
What's next for ICUP 👴🧓
We hope to find a better (possibly ML-based) inpainting algorithm, to reduce the size of the image artifacts left behind by the ex