☀️ 4D Neural Relightable Humans ☀️

👉Relighting4D: free-viewpoints relighting of humans under unknown illuminations

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅Relight dynamic, free viewpoints
✅Disentangled reflectance/geometry
✅SOTA on synthetic/real datasets
✅Code/models under MIT License

More: https://bit.ly/3RF3yH9

🍰 Long-Term Object Segmentation 🍰

👉XMem: object segmentation for long clips with unified feature memory stores

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅Inspired by Atkinson–Shiffrin model
✅Stores with different temporal scales
✅Memory consolidation algorithm
✅Compact/powerful long-term memory
✅Source code and models available

More: https://bit.ly/3PP0EOn

🔥🔥 Update 🔥🔥

👉Code https://github.com/THUDM/CogVideo

👉Demo https://wudao.aminer.cn/cogvideo/

More: https://bit.ly/3yP86BQ

🔥Grand Unification of Object Tracking🔥

👉UNICORN: unified method for SOT, MOT, VOS, & MOTS with a single neural net. 🤯

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅Great unification for 4 tracking tasks
✅Bridging methods / pixel-wise corresp.
✅SOTA on 8 challenging benchmarks
✅Source code under MIT License

More: https://bit.ly/3o74h6g

🔥OmniBenchmark: CV beyond ImageNet🔥

👉 21 realms, 7,000+ concepts and 1M+ images. Far beyond ImageNet!

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅vs. ImageNet: 2.5x realms, 9x concepts
✅Conciseness: no concept overlapping
✅ReCo: Relational Contrastive Learning
✅New supervised contrastive learning SOTA

More: https://bit.ly/3RJRKU0

💣 HD Neural Avatar @130FPS 💣

👉Samsung unveils MegaPortraits: novel one-shot creation of HD neural human avatar

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅One-shot neural avatars, SOTA up 512p
✅"Upgrading" to megapixel via more pics
✅First Neural Head Avatars in HD
✅Up to to 130 FPS via #GPU

More: https://bit.ly/3oboWWT

🔥Important update from #OpenAI🔥

👉 https://openai.com/blog/reducing-bias-and-improving-safety-in-dall-e-2/

🦚 TimeLens++: Event-based Interpolation 🦚

👉Novel event-based interpolation with non-linear flow & multi-scale fusion

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅Novel motion spline estimator
✅Non-linear continuous event/frames flow
✅Multi-feature fusion, gated compression
✅Novel hybrid dataset with 100+ videos

More: https://bit.ly/3yJyY6g

🪰NUWA-Infinity is out!🪰

👉∞ generation by #Microsoft: arbitrarily-sized HD images and long videos 🤯

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅Unconditional Image Gen.
✅Text-to-Image/Text-to-Clip
✅Animation / Out-painting
✅Hi-res, arbitrary long clip
✅NCP for patches caching

More: https://bit.ly/3zmBf9f

🔥 #AIwithPapers: we are 3,500+! 🔥

💙💛 Ready for YOLO 10, 11, π, ∞, Ψ, and more? The more we are, the faster we catch'em all 💙💛

😈 Invite your friends -> https://t.iss.one/AI_DeepLearning

🎷🎷OMNI3D: #3D Objects in the Wild🎷🎷

👉#3D detection: 234k images, 3M+ instances & 97 categories

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅OMNI3D from publicly released dataset
✅234k pics, 3M+ annotation with 3D box
✅97 categories such as sofa, table, cars
✅Fast (450x) and exact algorithm for IoU
✅Cube R-CNN: novel 3D object detector

More: https://bit.ly/3cznjzG

👹Multiface Neural Rendering 👹

👉A new multi-view, Hi-Res data collected at #META Reality Labs for neural face

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅Mugsy, large scale multi-cam apparatus
✅High-Res sync facial performance
✅Closing the gap in accessing HQ data
✅Suitable for #VR & #mixedreality

More: https://bit.ly/3b6XfeL

💄DEVIANT: SOTA in mono-3D detection💄

👉A novel Depth EquiVarIAnt NeTwork for 3D monocular detection in the wild

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅Michigan + #Meta + Ford 🤯
✅Depth-equi. + scale equiv. steerable
✅New SOTA on KITTI & Waymo
✅Ok cross-dataset -> generalization

More: https://bit.ly/3OEFtgK

🧱 Assembling #LEGO with #AI 🧱

👉Step-by-step assembly manual created by human into machine-interpretable instructions

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅Stanford + MIT + #Google 🤯
✅MEPNet: Manual-to-Executable-Plan Net
✅Manual to machine-executable plan
✅2D manual - 3D geometric shape
✅Reasoning on 3D alignments of legos

More: https://bit.ly/3PCwn5C

🔥🔥 UPDATE 🔥🔥

Code Released: https://github.com/apple/ml-mobileone

🎃New SOTA in UDA Semantic Seg.🎃

👉HRDA: multi-res Unsupervised Domain Adaptive Semantic Seg. -> SOTA

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅ETH + MPG + KU Leuven 🤯
✅HRDA: multi-res approach for UDA
✅Manageable GPU memory footprint
✅Small objects & fine segmentation detail
✅New SOTA on GTA and Synthia dataset

More: https://bit.ly/3cKtDEp

⚗️ SemAbs: 3D Scene Understanding ⚗️

👉Framework that equips 2D Vision-Language Models (VLMs) with new 3D spatial capabilities

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅2D VLMs with 3D reasoning skills
✅ViTs Efficient MS Relevancy Extraction
✅Novel Open-World understanding tasks
✅Completing partially observed objects
✅Finding hidden objects from language

More: https://bit.ly/3PYYk7d

🦚 TinyCD: Neural Change Detection 🦚

👉TinyCD: new SOTA in change detection with up to 150x fewer parameters.

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅SOTA with up to 150X fewer params
✅Mixing blocks for s.t. cross-correlation
✅PW-MLP for pixel wise classification
✅MAMB: novel block for skip connection

More: https://bit.ly/3zFEngk

🦊 3D-Aware "StyleGANv2" version 🦊

👉Upgrading StyleGANv2 into a novel 3D-aware GAN with just a minimal set of changes🤯

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅MPI-like 3D-aware GAN w/ single-view
✅GMPI: generative multiplane image
✅2D GAN 3D-aware with a minimal changes
✅Encoding 3D-aware inductive biases

More: https://bit.ly/3OJ5gnS

📺 NeRF-ing "The Big Bang Theory" 📺

👉Berkeley unveils an approach for accurate estimation of actor’s 3D pose & location

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅Input: images across the whole season
✅3D context (i.e. cams, structure, body)
✅Integrating context in 3D estimation
✅Re-ID, gaze, cinematography, pic editing
✅Knock, Knock, Penny!

More: https://bit.ly/3OLuaUb