💎 SJC: from 2D -> #3D generation 💎

👉 Novel method to convert pretrained 2D diffusion into 3D generative mode

😎Review https://bit.ly/3UppHJx
😎Paper arxiv.org/pdf/2212.00774.pdf
😎Code github.com/pals-ttic/sjc
😎Project pals.ttic.edu/p/score-jacobian-chaining

🤪 Just another ORDINARY week in #AI community 🤪

😈 Let's talk -> https://bit.ly/3VsEfth
😈 Invite friends -> https://t.iss.one/AI_DeepLearning

🎨 Dr.3D: adapting 3D GANs to Arts 🎨

👉Novel 3D GAN domain adaptation method for drawings. From portraits to 3D!

😎Review https://bit.ly/3He3Phv
😎Paper arxiv.org/pdf/2211.16798.pdf
😎Project jinwonjoon.github.io/dr3d/
😎Code github.com/JinWonjoon/Dr.3D

🫐 Tracking EVERYTHING in the wild 🫐

👉Disentangling classification from tracking, Track Every Thing Accuracy (TETA) for SOTA

😎Review https://bit.ly/3F2rfU7
😎Paper arxiv.org/pdf/2207.12978.pdf
😎Project www.vis.xyz/pub/tet/

👽HD Faces via "Hacked" Smartphone👽

👉A novel method for HD faces from (polarizing-foils) smartphones

😎Review https://bit.ly/3H8yhJN
😎Paper arxiv.org/pdf/2212.01160.pdf
😎Project dazinovic.github.io/polface

💎 Clothes-Changing Re-Identification 💎

👉The new SOTA in wild Re-identification by mixing appearance & face features

😎Review https://bit.ly/3HdTTUU
😎Paper arxiv.org/pdf/2211.13807.pdf
😎Project www.vision.huji.ac.il/reface
😎Code github.com/bar371/ReFace

👘 SOTA: One-Shot Animatable Avatars 👘

👉ELICIT: synthesis & avatar animation -> NeRF representation in one-shot

😎Review https://bit.ly/3P3eZY6
😎Paper arxiv.org/pdf/2212.02469.pdf
😎Project https://elicit3d.github.io/
😎Code github.com/huangyangyi/ELICIT

🐸 RANA: Neural Avatar by #Nvidia 🐸

👉Neural humans under arbitrary viewpoints, body poses & lighting

😎Review https://bit.ly/3iIYOTP
😎Paper arxiv.org/pdf/2212.03237.pdf
😎Project nvlabs.github.io/RANA

🧰UDE: Unified Human Motion Generation🧰

👉The first unified human motion from Text or Audio. Audio ON, New SOTA.

😎Review https://bit.ly/3HjWsVN
😎Paper arxiv.org/pdf/2211.16016.pdf
😎Project zixiangzhou916.github.io/UDE
😎Code github.com/zixiangzhou916/UDE

🔥 #StableDiffusion 2.1 is out! 🔥

👉Stability.ai releases Stable Diffusion v2.1 and DreamStudio-Updates_7-Dec 22.

😎Review https://bit.ly/3Y49898
😎Code github.com/Stability-AI/stablediffusion
😎Blog stability.ai/blog/stablediffusion2-1-release7-dec-2022

🪂 Fast Neural Deferred Shading 🪂

👉Novel fast multi-view 3D reconstruction with arbitrary materials & light

😎Review https://bit.ly/3Brs9bY
😎Project fraunhoferhhi.github.io/neural-deferred-shading
😎Code github.com/fraunhoferhhi/neural-deferred-shading
😎Paper mworchel.github.io/assets/papers/neural_deferred_shading_with_supp.pdf

🦄 #3D Human Motion from Speech 🦄

👉The new SOTA in generating 3D holistic body motions from human speech

😎Review https://bit.ly/3W59RVB
😎Project talkshow.is.tue.mpg.de/
😎Paper arxiv.org/pdf/2212.04420.pdf

💎4K-NeRF: Ultra High Resolution NeRF💎

👉Alibaba unveils the novel SOTA in Ultra High Resolution NeRF -> Up to 4K!

😎Review https://bit.ly/3hh3YGj
😎Code github.com/frozoul/4K-NeRF
😎Paper arxiv.org/pdf/2212.04701.pdf

👁️ Eye Tracking: the Tear Film Lipid 👁️

👉 Berkeley unveils a novel approach for DED, fully CV-based

😎Review https://bit.ly/3FndGyY
😎Project easytear-dev.github.io/
😎Paper arxiv.org/pdf/2212.03450.pdf

🐲 Neural Rendering from Web Images 🐲

👉Snap opens the novel SOTA in neural rendering from online image collections

😎Review https://bit.ly/3VPOjwy
😎Project formyfamily.github.io/NeROIC/
😎Paper arxiv.org/pdf/2201.02533.pdf
😎Code github.com/snap-research/NeROIC

🔥 E2POSE: code is out! 🔥

👉E2Pose: fully convolutional networks for e2e multi-person pose estimation

😎Review https://bit.ly/3WelGsP
😎Code github.com/AISIN-TRC/E2Pose
😎Tutorial cedro3.com/ai/e2pose/
😎Paper, not yet 🤷‍♂️

🧙Distortion-Aware Neural Selection 🧙

👉#Adobe unveils a novel method for interactive workflows used for decaling, texturing, or painting on surfaces

😎Review https://bit.ly/3uSou3b
😎Project threedle.github.io/DA-Wand
😎Paper https://drive.google.com/file/d/1OkGwIPQb9RUeT5aw-X_tdBAPG0bB4dNQ/view

🔥 GhostNetV2: Pure Neural Fire! 🔥

👉 Novel HW-friendly attention & GhostNetV2 for mobile. Top-1 with only 167M FLOPs

😎Review https://bit.ly/3j2EIUJ
😎Paper arxiv.org/pdf/2211.12905.pdf

🧲 MAGVIT: 250x Faster Transformer 🧲

👉MAGVIT: ONE model -> 10+ video synthesis tasks!

😎Review https://bit.ly/3FX2P09
😎Project magvit.cs.cmu.edu/
😎Paper arxiv.org/pdf/2212.05199.pdf
😎Code github.com/MAGVIT/magvit

🎯 MOTRv2: Tracking by Pretrained Model 🎯

👉MOTRv2: e2e multi-object tracking with pretrained detector

😎Review https://bit.ly/3UYzvdN
😎Paper arxiv.org/pdf/2211.09791.pdf
😎Code github.com/megvii-research/MOTRv2