🪩 DISCO: Human Dance Generation 🪩

👉NTU (+ #Microsoft) unveils DISCO: a big step towards the Human Dance Generation.

😎Review https://t.ly/cNGX
😎Paper arxiv.org/pdf/2307.00040.pdf
😎Project disco-dance.github.io/
😎Code github.com/Wangt-CN/DisCo

🛣️ STAR.: 3D-tracking w/ attention paradigm 🛣️

👉#Mercedes STAR: e2e 3D object tracking that follows the tracking-by-attention paradigm

😎Review https://t.ly/JoGj
😎Paper arxiv.org/pdf/2306.17602.pdf
😎Project simondoll.github.io/publications/star_track

🍡 Text2Cinemagraphs: Cinemagraph from text 🍡

👉CMU (+ #Snap) unveils a fully automated method for creating cinemagraphs from text descriptions

😎Review https://t.ly/BwZs6
😎Paper arxiv.org/pdf/2307.03190.pdf
😎Project text2cinemagraph.github.io/website
😎Code github.com/text2cinemagraph/text2cinemagraph

🔥Test-Time Training on fire 🔥

👉Extending the TTT to the streaming setting. Suitable for Panoptic, Instance & Colorization.

😎Review https://t.ly/eZYA
😎Paper arxiv.org/pdf/2307.05014.pdf
😎Project https://video-ttt.github.io/
😎Code github.com/renwang435/video-ttt-release

🃏 Deepfake via casual self-scan 🃏

👉TAU presents a novel approach to reenact an ID using only a casual self-scan

😎Review https://t.ly/9T8Wi
😎Paper arxiv.org/pdf/2307.06307.pdf
😎Project arielazary.github.io/PGR

🎪 Extreme Human Pose Estimation 🎪

👉RePoGen: novel synthetic data generator of extreme/realistic poses of humans

😎Review https://t.ly/ecBvM
😎Paper arxiv.org/pdf/2307.06737.pdf
😎Project mirapurkrabek.github.io/RePoGen-paper
😎Code github.com/MiraPurkrabek/RePoGen

💡 DATID-3D: Text-to-3D Generation 💡

👉 A novel domain adaptation method for 3D via text-to-image diffusion. 🤗-Demo available!

😎Review https://t.ly/TCL-B
😎Paper arxiv.org/pdf/2211.16374.pdf
😎Project gwang-kim.github.io/datid_3d/
😎Code github.com/gwang-kim/DATID-3D
🤗 huggingface.co/spaces/gwang-kim/DATID-3D
😎Colab colab.research.google.com/drive/1e9NSVB7x_hjz-nr4K0jO4rfTXILnNGtA?usp=sharing

🧯Neural Focal Modulation VAR🧯

👉A novel architecture for video recognition that models both local/global context

😎Review https://t.ly/rF_fk
😎Paper arxiv.org/pdf/2307.06947.pdf
😎Project talalwasim.github.io/Video-FocalNets
😎Code github.com/TalalWasim/Video-FocalNets

🐈 Gen-AI as representation learner 🐈

👉DreamTeacher: novel self-supervised feats. representation learning framework that utilizes gen-nets for pre-training downstream image backbones

😎Review https://t.ly/RL8iG
😎Paper arxiv.org/pdf/2307.07487.pdf
😎Project research.nvidia.com/labs/toronto-ai/DreamTeacher

☔ #SelfDriving? It's all about weather! ☔

👉Novel self-supervised MDE method to handle adverse weather in real-world autonomous driving

😎Review https://t.ly/tcLQW
😎Paper arxiv.org/pdf/2307.08357.pdf
😎Project kieran514.github.io/Robust-Depth-Project/

🦙 Llama-2: the Open-Source "ChatGPT" 🦙

👉GenAI, #Meta unveils Llama-2: a collection of LLMs ranging in scale 7-70B params. Challenging with #chatgpt, but open.

😎Review https://t.ly/bLJgP
😎Paper https://t.ly/AOXru
😎Project https://ai.meta.com/llama

🍉 AltFreezing: new SOTA in detecting deepfake 🍉

👉#Microsoft unveils AltFreezing: spatial/temporal artifacts in one model for more general face forgery detection

😎Review https://t.ly/mkIKX
😎Paper https://t.ly/z4KnJ
😎Code github.com/ZhendongWang6/AltFreezing

🪟META's Ultra-HD Data for #AR🪟

👉Aria Digital Twin: egocentric dataset for detection/tracking, reconstruction/understanding, S2R learning, pose and more.

😎Review https://t.ly/MRPt1
😎Paper arxiv.org/pdf/2306.06362.pdf
😎Project www.projectaria.com/datasets/adt
😎Code github.com/facebookresearch/projectaria_tools

👩‍🦰 Ultra-Realistic Neural Hair 👩‍🦰

👉A novel method to reconstruct the hair geometry at a strand level from monocular video or multi-view images

😎Review https://t.ly/6xZyp
😎Paper arxiv.org/pdf/2306.05872.pdf
😎Project samsunglabs.github.io/NeuralHaircut
😎Code github.com/SamsungLabs/NeuralHaircut

💪 Muscles in Action with #AI 💪

👉Muscles in Action (MIA): learn to incorporate muscle activity into human motion representations

😎Review https://t.ly/hUKub
😎Paper arxiv.org/pdf/2212.02978.pdf
😎Project musclesinaction.cs.columbia.edu

🪤 PAPR: Proximity Attention Point Render 🪤

👉PAPR: fast point-based scene representation with differentiable renderer approach

😎Review https://t.ly/yoI0g
😎Paper arxiv.org/pdf/2307.11086.pdf
😎Project https://zvict.github.io/papr

🪛 CAD-based Object Segmentation 🪛

👉 A novel three-stage approach to segment unseen objects in RGB images using their CAD models

😎Review https://t.ly/RtHLN
😎Paper arxiv.org/pdf/2307.11067.pdf
😎Code https://github.com/nv-nguyen/cnos

🛵 ALPR via CTS-Matching 🛵

👉UIT unveils a neural approach (#YOLO5 + tracking + rotation) to improve the license plate recognition accuracy

😎Review https://t.ly/VP4BP
😎Paper arxiv.org/pdf/2307.11336.pdf
😎Code github.com/chequanghuy/Character-Time-series-Matching

🥬 Generative AI’s Next Frontiers 🥬

👉Hair simulation, 2D->3D animation, and much more. ~20 papers from #NVIDIA accepted into #SIGGRAPH2023

😎 Review https://t.ly/wgGin

🦀 simPLE: learning to grasp only with CAD 🦀

👉simPLE learns to pick, regrasp & place objects precisely, given only the object CAD model and no prior experience

😎Review https://t.ly/ab5pA
😎Paper arxiv.org/pdf/2307.13133.pdf
😎Project mcube.mit.edu/research/simPLE.html