RenderDiffusion: 3D-aware Diffusion #Adobe unveils the first 3D-aware diffusion model purely trained from 2D images.

Bias in #AI: explained simple I asked DallE-mini on Hugging Face spaces (damn, you rock \m/) to generate images from simple sentences. Results are pretty clear and explain what the BIAS is without any word. Maybe the society (where data comes from) is. AI just (usually) learns from examples Want more?

E2E Unified Scene Text/Layout Detection #Google Research opens the world's first hierarchical scene text dataset a novel detection method.

Novel pre-training strategy for #AI EPFL unveils a novel pre-training strategy, MultiMAE: Multi-modal Multi-task Masked Autoencoders?

📫📫gDNA: Generative Detailed Neural Avatars📫📫 👉Detailed generation of 3D shapes of people in a variety of garments with corresponding skinning weight...

🔥🔥Transformer-based object tracking🔥🔥 👉ETH Zurich unveils a tracker architecture employing a Transformer-based model prediction module. 𝐇𝐢𝐠𝐡...

StyleGAN-XL: SOTA in neural image synthesis StyleGAN sets new standards for generative modeling but performance severely degrades on unstructured datasets (ImageNet). Researcher from Tübingen unveil StyleGAN-XL, new SOTA for large diverse dataset?

P2PNet: SOTA in crowd analysis is INSANE Localizing individuals in crowds is more than simply counting people. Tencent unveils P2PNet to directly predict heads in images with a deeper analysis.

Just damn #AI I reviewed hundreds of papers, checked as many repo, answered thousands of public/private messages. A small #AI community spontaneously grew up here: 15,000 people are now following my posts on LinkedIn. Something that I never expected and that flatters me. To thank you all I'm opening my private #Telegram group with 300 reviews, daily updated: (almost) every posts I have written from 2017 to date. A "repository" to dive in the #AI world and to keep in touch in real time.

🐯🐯Data2vec: learning everything with NO labels🐯🐯 👉Much of #AI is still based on supervised learning, which works exclusively with labeled data. #...