Taken on an abnormally sunny January weekend on my way to the subway.

A little hard to see, but here’s the bike and location in Google Maps

CV Dazzle: ITP Thesis Demo Look #1 (after) from Adam Harvey on Vimeo.

A few new visualizations I made to show how CV Dazzle techniques can fool face detection. Looks #2 and #3 are from a collaboration with DIS magazine from 2010.

• with DIS Magazine: Look #2 (before)
• with DIS Magazine: Look #2 (after)
• with DIS Magazine: Look #3 (before)
• with DIS Magazine: Look #3 (after)
• ITP Thesis Demo Look #1 (before)
• ITP Thesis Demo Look #1 (after)

Credits

• DIS Magazine
• Hair: Pia Vivas
• Model: Jen

OpenCV Visualization: Fronaltface Default Cascade from Adam Harvey on Vimeo.

Visualization of the detector shapes used in the “frontalface_default.xml” cascade file for OpenCV.

Each cascade can have 20 or so stages. The beginning stages are easier to pass, as noted by the larger shapes. As the algorithm progresses through each stage, some image subregions are quickly rejected. This allows the algorithm to quickly scan the image and not waste on unimportant areas. While towards the end of the visualization, the shapes will become smaller, more complicated and more picky.

The Viola Jones algorithm compares the difference between the dark and light areas defined by the shapes and measures whether this difference falls within a threshold.

More visualizations:

• Frontalface Alt Cascade
• Frontalface Alt2 Cascade
• Frontalface Alt Tree Cascade
• Eye Cascade

Image I made for Artstars*, because it keeps me going.

Based on the painting of Nadja from Carole Freeman, which I really liked.
EDIT: Found more of Carole Freeman’s work. Especially like this series: If Paparazzi Could Paint.