ABDELLATIF BELMADY’s Post

Introducing YOLO-World: The Future of Open-Vocabulary Object Detection Object detection is crucial for AI systems operating in the real world, but traditional models are limited to predefined object categories. YOLO-World is a revolutionary new model that overcomes this through open-vocabulary object detection, allowing it to identify any object described in natural language. To learn about it in detail read https://lnkd.in/dxtEPaXS The key innovations behind YOLO-World include: ⚡ Building on the exceptional speed and efficiency of YOLOv8 🪄 Using a "prompt-then-detect" paradigm to bypass real-time text encoding 🧠 Grounding visual and language understanding through cross-modal fusion ⚡ Avoiding complex Transformer architectures for faster, lighter inference 🌎 Released open-source under GPL to foster accessibility and collaboration YOLO-World achieves remarkable performance benchmarks, running up to 20x faster than comparable models on a single GPU while maintaining competitive accuracy. Its small model footprint enables deployment across a wide range of devices. But what really unlocks YOLO-World's transformative potential is custom prompting. With phrases like "cracked smartphone screen", you can rapidly adapt its detection capabilities without retraining. This opens up endless innovative applications: 📷 Intelligent surveillance and anomaly detection 🏪 Retail analytics and automated inventory tracking ⚙️ Enabling smarter robotics and navigation 👩🦯 Advancing accessibility for the visually impaired ...and so much more! YOLO-World heralds a new era of open-vocabulary, real-time perception for AI. Its speed, flexibility, and open-source philosophy empower us to build truly intelligent systems attuned to the real world's boundless complexity. #ai #computervision #innovation #opensource

"YALTAi introduces a game-changing approach to layout analysis in OCR and similar tasks. By leveraging object detection instead of pixel classification, it significantly enhances segmentation efficiency. The incorporation of YOLOv5 into Kraken 4.1's pipeline yields remarkable performance gains, particularly on smaller datasets. This innovation marks a pivotal shift in document digitization, promising superior extraction accuracy and noise reduction. #AI #MachineLearning #ComputerVision #YALTAi #KrakenEngine"

***Research Paper/Project Update*** Excited to share our recent work accepted at ECCV 2024 spearheaded by Abrar Majeedi 😁 🎉. Do checkout this work that he plans to present in Milan. One line description about project: How to assess the performance of a diver (or actions of a surgeon) and score it? How certain can we be about the model's predicted score and can we have a step-wise feedback to trust the model's predictions? Fun coincidence: One of the datasets we evaluated is related to Olympics Diving and the timing of this paper with 2024 Paris Olympics felt nice 😎 #MachineLearning #AI #ComputerVision #SportsResearch #ECCV

🚨 New Study Reveals Critical Blind Spot in AI's Visual Intelligence! 🚨 A recent study exposed a critical blind spot in AI's visual intelligence. While these models excel at interpreting photographic realism, they falter dramatically when presented with abstract visualizations like diagrams and charts. Advanced models such as GPT-4.0 and Claude 3.5 Sonnet achieved a mere 64.7% and 59.9% accuracy respectively on a diverse dataset of 11,193 abstract images, significantly lagging human performance at 82.1%. The research team employed a "multimodal self-instruct" approach to develop a diverse dataset of 11,193 abstract images, each accompanied by specific questions across eight scenarios, such as dashboards, road maps, diagrams, tables, and more. Despite their multimodal capabilities, the models struggle particularly with tasks requiring spatial reasoning and abstract concept understanding. Attempts to improve open-source model performance using synthetic data showed some success, yet reliance on costly closed models for reference data poses challenges. . 🔗Read more:- https://lnkd.in/dKbVgGzn. 🔗Blog :- https://t.ly/QLhNK #AI #ArtificialIntelligence #VisualIntelligence #MachineLearning #Research #DataScience #TechNews #DeepLearning #NeuralNetworks #AIResearch #TechInnovation #AbstractReasoning #AIChallenges #FutureOfAI #TechTrends #Innovation #SyntheticData #AIModels #Share

Hugging Face writes about a new transformer based AI model for predicting time-series data based on Meta's large language model Llama called Lag-Llama https://lnkd.in/eU89MeBJ. #huggingface #timeseriesforecasting #transformermodels #lagllama #llama #artificialintelligence #meta

Hugging Face Researchers Introduce Idefics2: A Powerful 8B Vision-Language Model Elevating Multimodal AI Through Advanced OCR and Native Resolution Techniques Quick read: https://lnkd.in/gHWWRkFr HF Page: https://lnkd.in/gJaAGSkd

Over the past years, foundation models have caused a paradigm shift in machine learning due to their unprecedented capabilities for zero-shot and few-shot generalization. However, despite the success in modalities such as natural language processing and computer vision, foundation models for time series forecasting have lagged behind. We are proud to present Lag-Llama: the first open-source foundation model for time series forecasting! Lag-LLama is developed jointly with our amazing team from Universite de Montreal, my CERC-AAI lab, ServiceNow and Morgan Stanley! We are grateful for compute resources provided to us by OLCF - the Oak Ridge Leadership Computing Facility (DoE Office of Science User Facility supported under Contract DE-AC05-00OR22725). Lag-Llama is a general-purpose foundation model for univariate probabilistic time series forecasting based on a decoder-only transformer architecture that uses lags as covariates. Lag-Llama is pretrained on a large corpus of diverse time series data from several domains, and demonstrates strong zero-shot generalization capabilities compared to a wide range of forecasting models on downstream datasets across domains. Moreover, when fine-tuned on relatively small fractions of such previously unseen datasets, Lag-Llama achieves state-of-the-art performance, outperforming prior deep learning approaches, emerging as the best general-purpose model on average. Lag-Llama serves as a strong contender to the current state-of-art in time series forecasting and paves the way for future advancements in foundation models tailored to time series data. Time-series foundation models have the potential to revolutionize applications such as computational medicine, natural sciences, finance climate, retail, ecology, energy and many more. They will become essential in applications where the data availability is limited and thus a transfer from a general-purpose pretrained model might be required. Model weights: https://lnkd.in/epuAwQ-X Demo: https://lnkd.in/evCHJ6-M GitHub: https://lnkd.in/eQi-uNzW Paper: https://lnkd.in/eK4BbfUk  Tweet: https://lnkd.in/ezBf8int Blog: https://lnkd.in/e6XFaNQS

Interesting news from the research team that developed an open source foundation model for time series which is at the state of the art in this sub-field. The paradigm shift for machine learning that foundation models triggered is opening to multiple use cases that in next months will go beyond the textual-language data and re-define the solutions in a number of industrial and practical contexts and time series are pervasive! See the list of training data set the team considered I propose into the first comment to have an idea of kind of application problems.

"Graph embeddings learn the structure of networks and represent it in low-dimensional vector spaces. Community structure is one of the features that are recognized and reproduced by embeddings. We show that an iterative procedure, in which a graph is repeatedly embedded and its links are reweighted based on the geometric proximity between the nodes, reinforces intra-community links and weakens inter-community links, making the clusters of the initial network more visible and more easily detectable. The geometric separation between the communities can become so strong that even a very simple parsing of the links may recover the communities as isolated components with surprisingly high precision. Furthermore, when used as a pre-processing step, our embedding and reweighting procedure can improve the performance of traditional community detection algorithms." https://lnkd.in/dfTk-HEJ.

If you're personally interested in AI ... Or if you're a *business* and you want to apply AI ... You need to understand how to use foundation models, and how to fine-tune them to solve specific problems for you. #machinelearning #datascience #AI #data