♻♻Koreans researchers and SAMSUNG SDS unveil ILVR, a novel generative process in DDPM for HQ images♻♻

👉Generative models, such as GAN, normalizing flows...

🎥💥Shanghai, 1929: #AI Restoration by Glamourdaze. Magic 🎥💥

🔈🔉🔊 SOUNDS ON 🔈🔉🔊

Highlights:
✅Cleaned noise artifacts
✅Increased framerate from 24 to…

☑☑ PySlowFast library by Facebook AI just added the neural Transformers support ☑☑

👉PySlowFast is an open source video understanding codebase from FAIR...

💥💥From #ICLR22, StyleNeRF: a novel #AI to synthesize HD images preserving 3D consistency💥💥

👉Photo-realistic free-view image synthesis of real-world... 16 comments on LinkedIn

🥝🥝UOAIS: a novel #AI to segment unseen objects via HOM scheme🥝🥝

👉Segmenting objects is an essential skill for robotic manipulations in an unstructured...

♻♻HARP: autoregressive latent video prediction #AI model for predicting High-Res future frames♻♻

👉Video prediction can enable agents to learn useful ...

🍧🍧Microsoft unveils the source code of Mesh Graphormer, a new Transformer-based #AI🍧🍧

👉3D human pose and mesh reconstruction from a single image ...

😵😵Huawei unveils HeadGAN, head reenactment, disentangling identity from expression😵😵

👉Recent attempts to solve the problem of head reenactment using... 15 comments on LinkedIn

🧁🧁CMU + NVIDIA open sourcing a novel self-supervised autoencoder for video generation🧁🧁

👉The art of seeing is in our ability to disentangle the reality...

🛸🛸 University of Zurich + Intel AI open-sourcing a framework for high-speed autonomous flight through challenging environments 🛸🛸

👉Quadrotors are... 12 comments on LinkedIn

🌳🌳NVIDIA and Microsoft unveil Megatron-Turing NLG 530B, the world’s most powerful generative language model🌳🌳

👉MT-NLG is the largest and the most... 11 comments on LinkedIn

🤯NVIDIA StyleGAN3 is out: source code available!🤯

👉Resolution and quality of images produced by GAN have seen rapid improvement recently. Despite their... 13 comments on LinkedIn

😵 New SOTA in video object segmentation is insane 😵

👉VOS aims at segment target instances in a video sequence. A high-performing #AI should be able... 127 comments on LinkedIn

🤪 The new SOTA in object tracking is awesome 🤪

👉Tracking by detection is the most effective paradigm for multi-object tracking in current. State-of-the-art... 11 comments on LinkedIn

😵😵𝑩𝒂𝒄𝒌 𝒕𝒐 𝒕𝒉𝒆 𝑭𝑬𝑨𝑻𝑼𝑹𝑬 with neural PixLoc. Source code available!😵😵

👉ETH Zurich & #Microsoft open-sourcing PixLoc, a neural network...

🧃🧃 A novel #AI paradigm for Video Instance Segmentation: Propose-Reduce just open-sourced 🧃🧃

👉Video instance segmentation (VIS) is a task to segment...

💎💎CMU is open-sourcing NeRS, a novel neural shape representation for surfaces💎💎

👉Recent history has seen the growth of implicit representations of...

🦑🦑 Non-Deep (∼10 layers) Neural Networks by Intel Labs are super-competitive 🦑🦑

👉Deep Neural Networks (DNNs) have revolutionized the fields of #AI...

👟👟❌❌ THIS SNEAKER DOES NOT EXIST ❌❌ 👟👟

Highlights:
✅Thousands of fake shoes via neural generation
✅Trained via #Nvidia StyleGAN2-ADA architecture
...

👽👽 #Nvidia just open-sourced DriveGAN, the most advanced neural simulator for #selfdrivingcars 👽👽

👉Realistic simulators are critical for training and… | 17 comments on LinkedIn