DINOv2 complements our other computer vision work, such as Segment Anything. While SAM is a promptable system focused on zero-shot generalization to diverse segmentation tasks, DINOv2 uses simple linear classifiers to achieve strong results across tasks beyond segmentation.
MULTIMODAL AI
-
ChatGPT Generates Ellsworth Kelly Style Abstract Art via JavaScript
By
–
Ellsworth Kelly sometimes used simple algorithms and pseudo-randomness to generate abstract art. I wonder what he'd make of ChatGPT creating an image in his style using JS, and employing Math.random() to generate a different color-scape each time. https://
jsfiddle.net/willknight/c2t
9sxpn/45/
… -
Bard vs Midjourney: AI Image Generation Comparison
By
–
les illustration ne sont a priori pas de magi, mais de bard… wait and see
-
Transformers: Six Years of Universal Neural Network Architecture
By
–
The "Attention is all you need paper" that introduced Transformer neural network architecture has been around for roughly six years. It's by far the first architecture to maintain its universality for a long time, not just for a single modality but for other modalities as well.
-
Zip-NeRF 3D rendering and Amazon Roomba mapping implications
By
–
Lots of chatter about what this video means for generative media. Zip-NeRF can render a 3D space quickly from a small collection of 2D images (with fewer errors). Also thinking about how this connects to Amazon buying Roomba, with all those maps of the inside of homes… https://t.co/37jloBNbkG
— Kate Crawford (@katecrawford) 16 avril 2023Lots of chatter about what this video means for generative media. Zip-NeRF can render a 3D space quickly from a small collection of 2D images (with fewer errors). Also thinking about how this connects to Amazon buying Roomba, with all those maps of the inside of homes…
-
Tactile Diffusion: Bridging Sim2Real Gap in Tactile Sensing
By
–
Tactile Diffusion generates synthetic tactile images from sim data, capturing the complex illumination of the gel deformation. This research from UW & Meta AI is the first method using diffusion to close the sim2real gap for vision-based tactile sensing.
— AI at Meta (@AIatMeta) 14 avril 2023
Read the paper ⬇️Tactile Diffusion generates synthetic tactile images from sim data, capturing the complex illumination of the gel deformation. This research from UW & Meta AI is the first method using diffusion to close the sim2real gap for vision-based tactile sensing. Read the paper
-
Stability AI Unveils SDXL Beta for Photorealistic Image Generation
By
–
Yesterday, we unveiled #SDXL beta, the latest Stable Diffusion model, which excels in photorealism and is slated for an open source release. SDXL is one of several foundation models from Stability AI that will be available on Bedrock!
-
Open-sourcing Animation Demo and 180K Drawing Dataset for Researchers
By
–
In 2021, we created a research demo that brought amateur drawings to life through animation — today, we're open-sourcing the code + releasing a first-of-its-kind dataset of nearly 180K annotated amateur drawings to help researchers keep innovating in this space.
— AI at Meta (@AIatMeta) 13 avril 2023
More details ⬇️In 2021, we created a research demo that brought amateur drawings to life through animation — today, we're open-sourcing the code + releasing a first-of-its-kind dataset of nearly 180K annotated amateur drawings to help researchers keep innovating in this space. More details
-
Game Development Using Stable Diffusion and AI Generation
By
–
How does this game work? I am assuming you are using stable diffusion for the skybox, but what about the shooting part?
-
Congratulations to Shinji W for Speech Research Excellence
By
–
Congratulations @shinjiw_at_cmu
, very well deserved ofc! Thank you for everything you’ve done and continue to, to push Speech research to the limits!