Spatial controllability is really hard for video gen, prompting will never do it alone. If a picture is worth 1000 words, a video is worth 24,000 words per second.
AI creative tools are going to be the biggest impact on creative work flows since the computer itself. we are all going to get amazing visual art, music, games, etc.
Billion params text-to-image models are amazing! But...not designed for editing real-world images/videos. Text2LIVE (ECCV oral) trains on 1 example and allows for various semantic, localized editing!
https://t.co/WYq2lI07YJ
@omerbartal@dolevofri @RafailFridman @yoni_kasten 1/3
Light Stage 3, @USC_ICT's 2002 system that surrounded actors with LED's to light them with images of virtual sets, is now on @ianfailes's Befores & Afters VFX Artifacts podcast series. I'll talk about Light Stage 3 and tons that's come after at @FMX_Conference tomorrow!
I encourage aspiring ML PMs to really learn the science, and read the textbooks. Leaving technical problem formulation entirely to the engineers/researchers is leaving so much of your impact value on the table.
The most important skill for AI product managers is ML problem formulation: the ability to express human-centric needs in terms of ML math that 1) actually exists and 2) actually solves the problem
#ML#Product#productmanagement#AI#design#MachineLearning
and 3) for which you can reasonably get or create training data. Human-centric design for AI often requires unusual/custom neural architectures you won't find in extant papers.And a harder time gathering/constructing peculiar datasets & supervision signals. But worth the effort!
Hello World! I am Herbert and I am a dinosaur of computer art. This year I will have my 95th birthday. My first (still analog) computer art originates from 1954. Much more art work followed till now ...
I am overwhelmed by your response! Thanks to all of you. Another image from 2009, titled flares - artistic visualisation of wavelets. My special interest: crossing borders between mathematics and visual arts.
A girl playing the Ukrainian national anthem in a bomb shelter during the shelling of Lysychansk, Luhansk Oblast, by the Russian troops on 6 March 2022.
📹@rescuero
20 years ago, we submitted a paper to @siggraph about surrounding actors with LEDs to light them with images of virtual environments, now a key component of virtual production stages being built and used around the world. #TBT@USC_ICT https://t.co/CDZOVACD4X
Some nice new work in natural image matting by @GoogleAI using a Light Stage and ratio matting to create training data of people in consistent lighting and backgrounds with ground truth alpha channels https://t.co/US1d0ddWFY @USC_ICT
Want to know the story of Deep Learning at Netflix? Read our paper in AI Magazine's Recommender Systems issue: "Deep Learning for Recommender Systems: A Netflix Case Study": https://t.co/tt0FrmnprR