AI with Papers - Artificial Intelligence & Deep Learning
15.2K subscribers
135 photos
247 videos
14 files
1.31K links
All the AI with papers. Every day fresh updates about #DeepLearning, #MachineLearning, LLMs and #ComputerVision

Curated by Alessandro Ferrari | https://www.linkedin.com/in/visionarynet/

#artificialintelligence #machinelearning #ml #AI
Download Telegram
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ¦„Time-Aware Neural VoxelsπŸ¦„

πŸ‘‰TiNeuVox: "NeRF" with time-aware voxel features 😡

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Dynamic scene w/ optimizable structure
βœ…Temporal information in radiance net
βœ…Small/large motion w/ single-res of feats
βœ…192Γ— faster than previous Hyper-NeRF

More: https://bit.ly/3wR4O08
πŸ‘11πŸ”₯2🀯1
🫐Neural Anomaly Detection by AWS🫐

πŸ‘‰Ultra-competitive inference and SOTA for both detection and localization

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Locally aggregated, mid-level feats patch
βœ…Maximizing nominal information at test time
βœ…Reducing biases towards ImageNet classes
βœ…Image-level anomaly AUROC of up to 99.6%

More: https://bit.ly/3t7Ndjg
πŸ”₯7🀯3πŸ‘2
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ›Ή Project Skate from Google #AI πŸ›Ή

πŸ‘‰#AI tool to analyze the skateboarder's tricks in real-time

More: https://bit.ly/3zbQS3M
πŸ”₯15🀩3πŸ‘1
This media is not supported in your browser
VIEW IN TELEGRAM
🧬Neural Text2Human Generation🧬

πŸ‘‰Text-driven neural human generation

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Full-body from a given human pose
βœ…Hierarchical texture-aware codebook
βœ…DeepFashion -> 44k Hi-Res images
βœ…Code and models available!

More: https://bit.ly/3Mdnpt0
πŸ”₯15πŸ‘1
🧨EfficientFormers: 1.6ms inference 🧨

πŸ‘‰Transformers fast as MobileNet? Snap shows that on #iphone!

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Low latency on mobile, high performance!
βœ…Revisiting the design of ViT through latency
βœ…New dimension-consistent design paradigm
βœ…EfficientFormers: a new ViT for mobile!

More: https://bit.ly/3MdgW15
πŸ”₯16πŸ‘1🀯1
This media is not supported in your browser
VIEW IN TELEGRAM
🐒 Transformer-Based Sens-Fusion 🐒

πŸ‘‰Updating TransFuser (CVPR21): image + LiDAR representations with self-attention

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Existing approach can't handle traffic 😒
βœ…Novel multi-modal fusion transformer
βœ…The new SOTA in driving performance
βœ…Reducing avg collisions per KM by 48%
βœ…Insights on current limitations of E2E

More: https://bit.ly/391dmd6
πŸ‘11πŸ”₯2
πŸ§˜πŸ»β€β™‚οΈYogNet: neural yoga assistantπŸ§˜πŸ»β€β™‚οΈ

πŸ‘‰Multi-person yoga neural expert for 20 asanas

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…CNNs & reg.LSTMs + 3D-CNNs
βœ…Multi-person asanas in real-time
βœ…YAR: dataset for yoga & posture
βœ…1206 videos, 2D RGB camera

More: https://bit.ly/3NncVbE
❀13πŸ‘1
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ”΄ Geogram: geometric algos in C++ πŸ”΄

πŸ‘‰Novel open-source programming library with (research) geometric algorithms in C++

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Geometry Processing from #INRIA
βœ…30+ papers from SIGGRAPH, etc.
βœ…Grants: GOODSHAPE & VORPALINE
βœ…Code (mostly C++) under BSD 3

More: https://bit.ly/3mhS4L7
πŸ”₯6πŸ‘3❀1
🍏 Open Source Vision from #Apple 🍏

πŸ‘‰CVNets: open-source (not a joke) lib for neural vision.

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…PyTorch-based neural lib. for vision
βœ…Train 2βˆ’4Γ— longer w/ augmentations
βœ…Plug-and-play components for CV
βœ…Source code under a custom license

More: https://bit.ly/39d1dSj
πŸ‘9
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ‡πŸ»Neural Clips by #Nvidia: INSANE πŸ‡πŸ»

πŸ‘‰Neural generation with changes in camera viewpoint & content that arises over time 🀯

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Novel hierarchical generator architecture
βœ…Temp. receptive field + temporal embed.
βœ…Multi-res. with super-resolution network
βœ…SOTA in long clip with motion & changes
βœ…Code, data & models in August 2022 πŸ–οΈ

More: https://bit.ly/3zroWsC
🀯9πŸ‘Ž2❀1
This media is not supported in your browser
VIEW IN TELEGRAM
⚽ Zero to #Messi with #deeplearning ⚽

πŸ‘‰EA unveils a neural system to learn multiple soccer juggling skills 😍

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Learning difficult soccer juggling skills
βœ…Layer-wise mixture-of-experts architecture
βœ…Specialization arises naturally
βœ…Adaptive random walk training strategy

More: https://bit.ly/3mwRaL2
πŸ”₯7πŸ‘3
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ–οΈ HumanNeRF: source code is out! πŸ–οΈ

πŸ‘‰Pausing the video at any frame and rendering the subject from arbitrary views!

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Synthesizing photorealistic humans
βœ…Synthesizing details, ie. cloth & face
βœ…Volumetric canonical T-pose
βœ…Skeletal rigid/non-rigid decomposition

More: https://bit.ly/3NEkTNY
🀯17πŸ”₯5πŸ‘2
This media is not supported in your browser
VIEW IN TELEGRAM
πŸŽ’ EG3D: source code is out! πŸŽ’

πŸ‘‰#Nvidia just opened EG3D: real time multi-view faces w/ HQ #3D geometry!

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Tri-plane-based 3D GAN framework
βœ…Pose-correlated attribute (expression)
βœ…SOTA in uncond. 3D-aware synthesis
βœ…Source code & models NOW available!

More: https://bit.ly/3aOfHs0
πŸ”₯7🀯6πŸ‘4❀2
πŸ”₯One Millisecond Backbone. Fire!πŸ”₯

πŸ‘‰MobileOne by #Apple: efficient mobile backbone with inference <1 ms on #iPhone12!

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…75.9% top-1 accuracy on ImageNet
βœ…38Γ— faster than MobileFormer net
βœ…Classification, detection & segmentation
βœ…Source code & model soon available!

More: https://bit.ly/3tsT7f2
❀24πŸ‘2
This media is not supported in your browser
VIEW IN TELEGRAM
🧨 Scaling Transformers to GigaPixels!🧨

πŸ‘‰Novel ViT called Hierarchical Image Pyramid Transformer (HIPT) -> Scaling to GigaPixels!

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Gigapixel whole-slide imaging (WSI)
βœ…Leveraging natural hier. structure of WSI
βœ…Self-supervised Hi-Res representations
βœ…Source code and models available!

More: https://bit.ly/3xLuzkg
🀯16πŸ‘1
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ‘—BodyMap: Hyper-Detailed HumansπŸ‘—

πŸ‘‰#META unveils 1st-ever dense continuous correspondence for clothed humans

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…1st-ever dense continuous corresp.
βœ…HQ fingers, hair, and clothes
βœ…Novel ViT-based architecture
βœ…SOTA on DensePose COCO

More: https://bit.ly/39nEPps
πŸ‘13❀2
🐹 NOAH just open-sourced! 🐹

πŸ‘‰A novel approach to find the optimal design of prompt modules through NAS algos.

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…NOAH from Neural prOmpt seArcH
βœ…Parameter-efficient β€œprompt modules”
βœ…Efficient NAS-based implementation
βœ…Better than transfer, few-shot & domain gen.

More: https://bit.ly/3MKfVhi
πŸ‘5πŸ‘2πŸ₯°1
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ„πŸ»β€β™€οΈNeural Super-Resolution in MoviesπŸ„πŸ»β€β™€οΈ

πŸ‘‰Implicit neural representation to get arbitrary spatial resolution & FPS -> Super Resolution!

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Video as continuous video representation
βœ…Clips in arbitrary space/time resolution
βœ…OOD generalization in space-time
βœ…Source code and models available

More: https://bit.ly/3xsqccf
πŸ”₯6πŸ‘2
This media is not supported in your browser
VIEW IN TELEGRAM
🧠 Bias in #AI, explained simple 🧠

πŸ‘‰Asking DallE-Mini to help me to show what the BIAS in #AI is

π†πžπ§πžπ«πšπ­πžπ π’πšπ¦π©π₯𝐞𝐬:
βœ…Best eng.->men/Caucasians
βœ…Best doctors->men/Caucasians
βœ…Top CEOs->men/Caucasians
βœ…Chef, kitchen->men/Caucasians
βœ…Rich People->only Caucasians
βœ…Poor People->non-Caucasians
βœ…Italian engineers->back in 30's
βœ…Chinese eng.->infrastructures
βœ…Italian working->local market
βœ…Chinese working->vegetables
βœ…Men workers->constructions
βœ…Women workers->only office

More: https://bit.ly/3b0UFqd
πŸ‘13❀6😁4
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ¦• SAVi++: Segmentation by #Google πŸ¦•

πŸ‘‰Novel unsupervised object-centric #AI to predict depth signals from slot-based video representation

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Segmenting complex dynamic scenes
βœ…Static/Moving objects on naturalistic BG
βœ…LiDAR-SAVi: segmenting in the wild
βœ…Source code and model soon available!

More: https://bit.ly/3n3hywd
πŸ”₯7πŸ‘6πŸ₯°1