Helpful tips from NIST on how to design anti-face recognition masks:
- Higher error rates in wide masks with high-coverage masks and black masks
- "A dark mask on light skin or a light mask on dark skin could cause overexposure problems"

Euclidean distance between the interpolative fanciness of OpenAI's DALL·E dream machines and the institutional dullness of yester-century's Columbia House/BMG compilations is approaching 0.0

Interesting because the same (seemingly) GAN generated faces also now available for free for any purpose here

For illustrative purposes IBM researchers used a GAN generated face in their Diversity in Faces paper, perhaps first ever GAN identity placeholder in research paper. Interestingly, they claim image is subject to license terms of algorithm author

Packet activism: Broadcast names of those who died trying to enter EU as locally visible network nodes by Christian Sievers

UPC: Learning Universal Physical Camouflage Attacks on Object Detectors

"We can also put fake heart rate and respiration in your video" - Rosalind Picard, Affectiva co-Founder

If it’s “breaking news” then maybe it’s actually newly received information. Otherwise it’s probably old news and therefore not news.

Interview with Ö1 Austria on Syrian Archive + VFRAME using computer vision to detect cluster munitions in conflict zones

WIRED: Facial recognition has already reached its breaking point
citing UnConstrained College Students dataset as one of many egregious examples. Read about UCCS dataset

In 2018 a researcher at Hikvision Research Institute used a collection of surveillance videos of students from Duke University campus to train their surveillance technologies

Thank you Ars Electronica for awarding VFRAME with a 2019 Award of Distinction in the Artificial Intelligence & Life Art category!

A month after the MegaPixels / Financial Times story on face recognition training datasets it appears that Microsoft's privacy-violating MS Celeb face recognition dataset is no longer available

Show older

Server run by the main developers of the project 🐘 It is not focused on any particular niche interest - everyone is welcome as long as you follow our code of conduct!