500 million+ members | Manage your professional identity. Build and engage with your professional network. Access knowledge, insights and opportunities.

 👗🥻👘 SenseTime open-sourced 𝐆𝐚𝐫𝐦𝐞𝐧𝐭𝟒𝐃, #AI for human-garment reconstruction👗🥻👘

👉Learning to reconstruct 3D garments is important for dressing...

😈👿 HeadNeRF: NeRF-based #AI for human faces 😈👿

👉Chinese researchers unveil a novel NeRF-based method to render Hi-Fi head images in real-time, and...

🍠🍠 ViT Deep features for Co-segmentation, Part Co-segmentation, and Semantic Correspondences 🍠🍠

👉“Deep Features” extracted from the activation of layers…

🥭🥭Real-ESRGAN: Super-Resolution with pure Synthetic Data🥭🥭

👉Single image super-resolution topic aims at reconstructing a high-resolution image from its… | 16 comments on LinkedIn

🎁🎁EgoBody: dataset for #3d human interactions🎁🎁

👉Understanding the social interactions from first-person view is crucial for robotics, #AR #VR and more…

🧁🧁EG3D: the new SOTA in #3D-aware GANs is insane!🧁🧁

👉Unsupervised high-quality multi-view consistent 3D images using only collections of single-view 2D… | 11 comments on LinkedIn

😮The Chinese #Musk doppelganger: #deepfake or… | 11 comments on LinkedIn

🤩🤩The SOTA in video instance segmentation is insane (every day better)🤩🤩

👉SeqFormer is the frustratingly simple model for video instance segmentation…

✂️🧴The #AI Barbershop source code is out! 🧴✂️

👉Even though recent work on GANs enables synthesis of realistic hair or faces, it remains difficult to… | 20 comments on LinkedIn

🗿🗿StyleSDF: the new SOTA in 3D-consistent image and shape generation🗿🗿

👉This #AI (Washington + Adobe + Stanford) is trained on single-view RGB data...

🐈🐈‍⬛#META unveils BANMo: 3D-neural animations from casual clips🐈‍⬛🐈

👉#3D shape reconstruction often relies on specialized sensors, or pre-built 3D deformable models (e.g., SMPL). Such methods cannot scale to diverse sets of objects in the wild. BANMo, by…

🐸🐸 StyleGAN-V: new SOTA in Continuous Video Generation is magic 🐸🐸

👉StyleGAN-V extends the recent paradigm of neural representations to the video generation domain. KAUST & #Snap show with this model that is possible to generate arbitrarily long clips at…

🤹‍♀️🤹‍♂️“Splice" #AI: painting the structure of an image with the appearance of another one🤹‍♀️🤹‍♂️

👉WAIC unveils an #AI to generate objects in a source structure image “painted” with the visual appearance of their semantically related objects in a target…

💃🕺HumanNeRF: pausing a clip at any time to render the subject💃🕺

👉A free-viewpoint rendering method of a human performing complex body motions (from #YouTube). This #AI enables pausing the video at any frame and rendering the subject from arbitrary new camera…

🍚🍚#Google open-sourced V-MoEs: Sparse Mixture of Experts for vision🍚🍚

👉MoEs have demonstrated excellent scalability in #NLP. Google #AI unveils the vision MoE->V-MoE, a sparse version of the Vision Transformer: scalable and competitive with the largest dense…

😮🤩 WebToonMe project made my day 😮🤩

- Webtoon Me: https://lnkd.in/dSi2JjvS
- Webtoon Painter: https://lnkd.in/d-2_Y3Bp
- Company: https://lnkd.in/dcVbemiz

#artificialintelligence #machinelearning #deeplearningai #deepneuralnetworks #neuralnetworks #ml …

😵‍💫🤪SeamlessGAN, endless texture generator, is hypnotic😵‍💫🤪

👉Real-time graphics requires HQ textured materials to convey realism in virtual env. Generating textures is challenging as they need to be realistic, seamlessly tileable, and low-memory.

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:…

🐸🐸#Nvidia: neural rendering in milliseconds🐸🐸

👉Neural graphics primitives, by fully connected NNs, are costly to train/evaluate. #Nvidia unveils a new input encoding that permits the use of smaller networks without sacrificing quality, with a speedup of…

🦚🦚ELSED: the best/fastest line detector ever🦚🦚

👉Detecting local features, such as corners, segments or blobs, is the first step in the pipeline of many #computervision applications. Its speed is crucial for real-time applications.

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
✅ELSED: Enhanced…