/* ---- Google Analytics Code Below */
Showing posts with label Facial Images. Show all posts
Showing posts with label Facial Images. Show all posts

Saturday, June 05, 2021

Robots Smiling Back

Not sure how much this buys you, but it makes sense where creating a android-like image to have emotion matching to group or task. We experimented only slightly with this in brand equity systems,  mostly keeping the image smiling.   It was suggested that we might include a quizzical or neutral face to indicate a more neutral position, but that was never integrated.

The Robot Smiled Back   By Columbia Engineering

Columbia Engineering researchers have spent five years developing an autonomous robot with a face that mimics the expressions of nearby humans.

Development of the EVA robot began with the construction of its physical mechanisms, engineered as a disembodied bust with a blue-colored face that can express basic, as well as subtler, emotions via artificial muscles.

The researchers used three-dimensional printing to manufacture complex components that could fit into EVA's human-sized skull without losing functionality. EVA employs deep learning artificial intelligence to "read," then imitate, expressions on nearby human faces, and learned to mimic faces by watching videos of itself.

The robot eventually became capable of reading human facial gestures from a camera, and of responding by mirroring those facial expressions.  .. "

Saturday, February 06, 2021

Where did Those Images come From?

Worked on an image recognition problem that could have used this approach to determine where certain images where coming from, either to validate them, or to determine which needed to be removed after attribution.   Only reflects images that have posted through flicker, which is limiting, but indicative.  Potential cleansing and risk analysis step. 

Here's a Way to Learn If Facial Recognition Systems Used Your Photos  By The New York Times, February 5, 2021

Part of a mosaic of about 50,000 images from the MegaFace dataset, which includes over 3.5 million.

A new online tool lets people search many image collections to see if their old photos were used to train facial recognition systems.

The Exposing.AI online tool lets people search image collections for old photos of themselves, in order to learn if such images were used to train facial recognition systems. The tool matches images from the Flickr online photo-sharing service, locating photos if users already have a way of pointing to them online, for example via an Internet address.

People can search only for images posted to Flickr, using a Flickr username, tag, or Internet address that can identify those pictures. The New York Times was able to use the tool to find photos that Exposing.AI indicated were used in facial recognition datasets.

From The New York Times   View Full Article - May Require Paid Subscription