Excited to share our work on #syntheticdata, generating dynamic expression-based wrinkles for digital faces, to appear at #WACV2023!
With @Friggog, @errollw, and Tadas Baltruลกaitis @Microsoft Mixed Reality & AI Cambridge
๐https://t.co/3oSjCBedHK
๐งต1/8
Very proud to share more about our work on ๐๐ฒ๐ป๐๐ฒ ๐๐ฎ๐ฐ๐ฒ ๐๐ฎ๐ป๐ฑ๐บ๐ฎ๐ฟ๐ธ๐, which will appear at #ECCV2022. Check it out! ๐
๐ https://t.co/aoqg2BVrNS
๐ฅ https://t.co/AL7fcDNPTv
๐ https://t.co/4fGrgxAYVF
I'm delighted to share our work on dense landmarks! https://t.co/4fGrgxAYVF
๐ Dense landmarks are all you need for accurate 3D face reconstruction and performance capture
๐ช No surprise: the secret sauce is synthetic data
๐ It's fast, running at 150FPS on a single CPU thread
Surprising and super valuable that you don't need Hollywood-quality synthetic faces to train DNNs that work on real data, allowing much more diverse training sets.
Turns out the uncanny valley is beyond the domain gap...