🤩🤩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…

🦎🦎GANmouflage: learn to non-detecting objects🦎🦎

👉GANmouflage by UMich & Berkeley learns to camouflage 3D objects within scenes. Given an object’s... 23 comments on LinkedIn

🧊🤖 GANcraft: Neural Photorealistic conversion of #Minecraft 🤖🧊

👉#NVIDIA and Cornell unveil GANcraft, neural rendering pipeline to generate realistic 3D…

👕👚Re-rendering human fashion style with GAN👚👕

👉VT + Adobe are open sourcing a new algorithm for generating detail-preserving and photo-realistic… | 19 comments on LinkedIn

▶️ ▶️ StylEx by #Google: novel visual explanation of #AI ▶️ ▶️

👉All the previous methods tend to change all attributes at once. Google proposes StylEx, a new…

🐯🐯Data2vec: learning everything with NO labels🐯🐯

👉Much of #AI is still based on supervised learning, which works exclusively with labeled data. #META (former #FacebookAI) unveils data2vec, a framework that uses the same learning method for either speech…

😈😈STIT: a novel GAN-based, temporal consistent, video facial editor😈😈

👉The ability of GANs to encode rich semantics is well-known. However, replicating...