AI with Papers - Artificial Intelligence & Deep Learning
15.2K subscribers
136 photos
248 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—ΊοΈNeural Translation Image -> MapπŸ—ΊοΈ

πŸ‘‰A novel method for instantaneous mapping as a translation problem

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Bird’s-eye-view (BEV) map from image
βœ…A restricted data-efficient transformer
βœ…Monotonic attention from lang.domain
βœ…SOTA across several datasets

More: https://bit.ly/39MQ76Z
πŸ”₯20πŸ‘6😱1
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ₯Ά E2V-SDE: biggest troll ever? πŸ₯Ά

πŸ‘‰E2V-SDE paper (accepted to #CVPR2022) consists of texts copied from 10+ previously published papers πŸ˜‚

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Latent ODEs for Irregularly-Sampled TS
βœ…Stochastic Adversarial Video Prediction
βœ…Continuous Latent Process Flows
βœ…More papers....


More: https://bit.ly/3bsL8Zw (AUDIO ON!)
πŸ‘9
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ”₯πŸ”₯YOLOv6 is out: PURE FIRE!πŸ”₯πŸ”₯

πŸ‘‰YOLOv6 is a single-stage object detection framework for industrial applications

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Efficient Decoupled Head with SIoU Loss
βœ…Hardware-friendly for Backbone/Neck
βœ…520+ FPS on T4 + TensorRT FP16
βœ…Released under GNU General Public v3.0

More: https://bit.ly/3OLjncK
πŸ”₯37πŸ‘6
This media is not supported in your browser
VIEW IN TELEGRAM
πŸͺ BlazePose: Real-Time Human Tracking πŸͺ

πŸ‘‰Novel real-time #3D human landmarks from #google. Suitable for mobile.

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…MoCap from single RGB on mobile
βœ…Avatar, Fitness, #Yoga & AR/VR
βœ…Full body pose from monocular
βœ…Novel 3D ground truth acquisition
βœ…Additional hand landmarks
βœ…Fully integrated in #MediaPipe

More: https://bit.ly/3uvyiAv
πŸ”₯14πŸ‘4
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ”₯YOLOv7: YOLO for segmentationπŸ”₯

πŸ‘‰YOLOv7: adding a lot of newer skills to the YOLO architecture family.

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…YOLOv7, not a successor of YOLO family!
βœ…Framework for detection & segmentation
βœ…Applications based on #META detectron2
βœ…DETR & ViT detection out-of-box
βœ…Easy support for pipeline thought #ONNX
βœ…YOLOv4 + InstanceSegm. via single stage
βœ…The latest YOLOv6 training is supported!
βœ…Source code under GPL license.

More: https://bit.ly/3ysSJAp
πŸ”₯22🀯9πŸ‘5😁2
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ”₯πŸ”₯ HD Dichotomous Segmentation πŸ”₯πŸ”₯

πŸ‘‰ A new task to segment highly accurate objects from natural images.

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…5,000+ HD images + accurate binary mask
βœ…IS-Net baseline in high-dim feature spaces
βœ…HCE: model vs. human interventions
βœ…Source code (should be) available soon

More: https://bit.ly/3ah2BDO
πŸ”₯13
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ”₯πŸ”₯ Neural Segmentation on fire πŸ”₯πŸ”₯

πŸ‘‰Novel methods for segmentation with mask calibration. Robustness++ in VOS.

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Study: VOS robustness vs. perturbations
βœ…Adaptive object proxy (AOP) aggregation
βœ…Less errors due unstable pixel-level match
βœ…Code/models (should be) available soon

More: https://bit.ly/3yhIY6Q
πŸ‘15❀1πŸ”₯1
This media is not supported in your browser
VIEW IN TELEGRAM
😊😎 Seq-DeepFake via Transformers 😎😊

πŸ‘‰S-Lab opens Seq-DeepFake: Detecting Sequential DeepFake Manipulation

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Seq-DeepFake: sequences of facial edits
βœ…Dataset: 85k #deepfake manipulation
βœ…Powerful Seq-DeepFake Transformer
βœ…Code, dataset and models available!

More: https://bit.ly/3ACQXhi
πŸ‘15πŸ”₯2❀1
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ¦’ Text2LIVE: Text-Driven Neural Editing πŸ¦’

πŸ‘‰#Amazon unveils a novel #AI for text-driven edit of videos. Insane! 🀯

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Semantic edits of real-world videos
βœ…Edit layer–RGBA representing target
βœ…Edit layers synthesized on single input
βœ…No masks or a pre-trained generator

More: https://bit.ly/3NVP6aE
🀯18πŸ‘9πŸ”₯8❀1
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“ŸπŸ“ŸAI-Designed Circuits with Deep RLπŸ“ŸπŸ“Ÿ

πŸ‘‰#Nvidia unveils an #AI to design circuits from scratch, smaller and faster than SOTA ones

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Parallel prefix circuits for Hi-Perf
βœ…RL framework to explore the circuit space
βœ…Smaller, Faster, Power-- from the scratch

More: https://bit.ly/3yY9dk7
🀯13πŸ‘5πŸ”₯3
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ‘½ Neural I2I with a few shoots πŸ‘½

πŸ‘‰#Alibaba unveils a novel portrait stylization. Limited samples (∼100) -> HD outputs

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Calibration first, translation later
βœ…Balanced distribution to calibrate bias
βœ…Spatially semantic constraints via geometry
βœ…Source code and models soon available!

More: https://bit.ly/3IwOmHO
❀10πŸ‘5😱1
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ€Ήβ€β™‚οΈ K-Means Mask Transformer πŸ€Ήβ€β™‚οΈ

πŸ‘‰#Google AI unveils kMaX-DeepLab, novel E2E method for segmentation

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…kMaX-DeepLab: k-means Mask Xformer
βœ…Rethinking relationship pixels / object
βœ…Cross-attention -> k-means clustering
βœ…The new SOTA on several dataset

More: https://bit.ly/3O2QV5I
πŸ”₯11πŸ‘2πŸ‘1
This media is not supported in your browser
VIEW IN TELEGRAM
β˜€οΈ 4D Neural Relightable Humans β˜€οΈ

πŸ‘‰Relighting4D: free-viewpoints relighting of humans under unknown illuminations

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Relight dynamic, free viewpoints
βœ…Disentangled reflectance/geometry
βœ…SOTA on synthetic/real datasets
βœ…Code/models under MIT License

More: https://bit.ly/3RF3yH9
πŸ”₯9πŸ‘2
This media is not supported in your browser
VIEW IN TELEGRAM
🍰 Long-Term Object Segmentation 🍰

πŸ‘‰XMem: object segmentation for long clips with unified feature memory stores

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Inspired by Atkinson–Shiffrin model
βœ…Stores with different temporal scales
βœ…Memory consolidation algorithm
βœ…Compact/powerful long-term memory
βœ…Source code and models available

More: https://bit.ly/3PP0EOn
🀯16πŸ‘5πŸ‘3
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ”₯Grand Unification of Object TrackingπŸ”₯

πŸ‘‰UNICORN: unified method for SOT, MOT, VOS, & MOTS with a single neural net. 🀯

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Great unification for 4 tracking tasks
βœ…Bridging methods / pixel-wise corresp.
βœ…SOTA on 8 challenging benchmarks
βœ…Source code under MIT License

More: https://bit.ly/3o74h6g
πŸ‘13πŸ”₯3🀯1😱1
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ”₯OmniBenchmark: CV beyond ImageNetπŸ”₯

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

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…vs. ImageNet: 2.5x realms, 9x concepts
βœ…Conciseness: no concept overlapping
βœ…ReCo: Relational Contrastive Learning
βœ…New supervised contrastive learning SOTA

More: https://bit.ly/3RJRKU0
πŸ”₯11🀩3
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ’£ HD Neural Avatar @130FPS πŸ’£

πŸ‘‰Samsung unveils MegaPortraits: novel one-shot creation of HD neural human avatar

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…One-shot neural avatars, SOTA up 512p
βœ…"Upgrading" to megapixel via more pics
βœ…First Neural Head Avatars in HD
βœ…Up to to 130 FPS via #GPU

More: https://bit.ly/3oboWWT
πŸ”₯22πŸ‘1πŸ‘1
This media is not supported in your browser
VIEW IN TELEGRAM
🦚 TimeLens++: Event-based Interpolation 🦚

πŸ‘‰Novel event-based interpolation with non-linear flow & multi-scale fusion

𝐇𝐒𝐠𝐑π₯𝐒𝐠𝐑𝐭𝐬:
βœ…Novel motion spline estimator
βœ…Non-linear continuous event/frames flow
βœ…Multi-feature fusion, gated compression
βœ…Novel hybrid dataset with 100+ videos

More: https://bit.ly/3yJyY6g
πŸ”₯16πŸ‘4