Research & Education

FSDrive is the official implementation for the research paper "FutureSightDrive: Thinking Visually with Spatio-Temporal CoT for Autonomous Driving," which was recognized as a NeurIPS 2025 spotlight. This tool introduces a novel spatio-temporal Chain-of-Thought (CoT) approach, allowing end-to-end autonomous driving Visual Language Agents (VLA) to visually process and plan trajectories. It uniquely unifies visual generation and understanding with minimal data, marking a significant advancement in applying visual reasoning to autonomous driving. FSDrive provides comprehensive instructions for installation, data preparation, training, inference, evaluation, and visualization, making it a valuable resource for researchers and developers in the autonomous driving domain.

Recommend Similar Papers is an AI tool hosted on Hugging Face Spaces that helps users discover related academic content. By simply entering the URL of a Hugging Face Papers entry, the application leverages Semantic Scholar’s recommendation service to provide a concise list of relevant papers. The tool returns the titles, publication years, and direct links to these papers, formatted in markdown for easy use. This makes it a valuable resource for researchers, students, and academics looking to quickly expand their literature review or find additional resources on a specific topic.

Explore Unitxt is an AI tool hosted on Hugging Face, offering a user-friendly interface for interacting with the Unitxt framework. This application is designed to facilitate various tasks, providing a platform for users to explore and utilize Unitxt's capabilities. While the specific functionalities are not detailed, the tool aims to simplify interaction with the underlying Unitxt system. It is free to use and operates as a web-based application, making it accessible to a broad audience interested in AI and task automation.

NobleAI offers a Science-Based AI and cloud-based VIP Platform designed to accelerate innovation in energy, chemistry, and manufacturing. By embedding scientific laws and domain knowledge into its models, NobleAI generates accurate, interpretable predictions even with limited data. The platform helps companies bring products to market faster, strengthen margins, and optimize product and asset performance, from formulation development to asset optimization. Key features include smart data and model management, insightful predictions, optimized designs, rich visualizations, and explainable AI. It supports diverse industries like energy, consumer packaged goods, and chemicals, enabling faster identification of sustainable alternatives and reducing trial-and-error in R&D.

Answer.AI is a comprehensive AI-powered platform designed to support students in all aspects of their academic journey. It functions as a 24/7 AI tutor and counselor, offering instant homework help, detailed step-by-step solutions for problems, and personalized study support. Beyond coursework, Answer.AI assists with college admissions guidance, AP tutoring, and even provides insights for managing social anxiety and collaborating with classmates. The platform includes features like flashcards, customizable quizzes, and study groups, tracking progress to ensure student success. Available as a mobile app and Chrome extension, Answer.AI aims to make academic support accessible to millions of students.

Budgerigar Gender Determination is an AI tool hosted on Hugging Face designed to automatically identify the gender of budgerigars. Users can upload photos or videos of their birds, and the application will analyze the cere color to determine gender. The tool then draws labeled boxes around each detected bird, indicating its gender. It offers adjustable confidence and detection settings, allowing users to fine-tune the analysis. This free tool provides a quick and easy method for budgerigar owners, bird enthusiasts, and researchers to determine the gender of their birds without manual inspection.

fpn.pytorch offers a pure PyTorch implementation of the Feature Pyramid Network (FPN) for object detection, building upon the properties of a faster R-CNN implementation. This project stands out for its complete conversion of all NumPy implementations to PyTorch, ensuring a consistent and efficient environment. A key feature is its support for training with batch sizes greater than one, achieved by revising all relevant layers including dataloader, RPN, and ROI-pooling. It also leverages a multiple GPU wrapper (nn.DataParallel) for flexible scaling across one or more GPUs. The implementation integrates three pooling methods—ROI pooling, ROI align, and ROI crop—all adapted for multi-image batch training. Benchmarking has been conducted on datasets like PASCAL VOC and COCO, demonstrating its performance.

AlgorithmicTrading is an open-source repository offering three distinct methods for identifying and exploiting arbitrage opportunities: Dual Listing Arbitrage, Options Arbitrage, and Statistical Arbitrage. Developed in collaboration with Optiver and peer-reviewed by their staff, this resource provides a robust foundation for understanding these complex financial strategies. While the analysis offers valuable insights into how these methods operate, the repository explicitly notes that effective implementation typically requires C++ for speed and a lightning-fast connection, making it less feasible for retail investors. It serves primarily as an educational and research tool for those interested in advanced algorithmic trading concepts.

Stardog offers a Semantic AI Platform utilizing an Enterprise Knowledge Graph to connect, model, and explore data across various sources like data lakes and warehouses. This platform helps enterprises and government agencies uncover hidden connections, accelerate insights, and power effective digital transformation. Key offerings include Stardog Cloud for a managed service, Stardog Voicebox for AI question-answering on enterprise data, and Stardog Explorer for visual data discovery. It also provides tools for ontology modeling, developer IDEs, and connectors for data connectivity, ensuring security and governance for compliant deployments. Stardog aims to empower data citizens with knowledge-informed decisions.

FacePose_pytorch provides a PyTorch implementation for real-time head pose estimation (yaw, roll, pitch) and emotion detection, boasting state-of-the-art performance. The tool is designed for easy deployment and use, offering high accuracy in solving various face detection problems. It utilizes Retinaface for face frame extraction, PFLD for key point identification, and a simple linear model for pose estimation. Additionally, it incorporates a highly accurate emotion recognition model, achieving impressive results on datasets like raf-db, affectnet, and ferplus, predicting seven types of expressions. The project emphasizes its efficiency and accuracy compared to existing open-source solutions.

Qonqur is a platform designed to liberate intellect and creativity by organizing research articles into interactive knowledge maps. It allows users to automatically organize dozens of articles by citation and knowledge dependencies, providing a clear view of where they are in their learning journey and where they are going. The tool supports self-study and research paths, and offers features like webcam gesture controls and advanced AI capabilities for deeper exploration. Qonqur aims to incorporate the arts into the sciences, offering a new paradigm for understanding and navigating complex information, from foundational concepts to frontier knowledge.

Pseudo_Lidar_V2 is an open-source project focused on advancing 3D object detection for autonomous driving by improving depth estimation. This tool, presented in an ICLR 2020 paper, builds upon the pseudo-LiDAR framework by enhancing stereo depth estimation, particularly for faraway objects. It also integrates sparse LiDAR sensor data to de-bias depth estimations through a proposed depth-propagation algorithm. The project provides code, pretrained models, and detailed instructions for training and inference on datasets like SceneFlow and KITTI, making it a valuable resource for researchers and developers in the autonomous driving domain.

This GitHub repository offers a Chinese translation of the second edition of the influential book "Reinforcement Learning: An Introduction." The project was initiated to provide a valuable resource for Chinese-speaking individuals interested in studying and discussing reinforcement learning concepts. While the project is now in maintenance mode due to the official Chinese translation being published, it still serves as a community-driven effort to make this complex topic more accessible. The repository includes translated chapters and aims to facilitate a deeper understanding of reinforcement learning algorithms and theories.

CogVLMv1 Captionner is an AI tool designed to generate detailed, factual descriptions of uploaded images. It identifies objects, analyzes backgrounds, and details other visual elements to provide a comprehensive caption. While the current live website indicates a runtime error, the tool's intended functionality is to offer users the ability to upload an image and, if desired, customize a prompt to guide the caption generation process, resulting in a tailored description. This makes it suitable for various applications requiring precise image analysis and textual representation.

V-JEPA 2 is an AI tool designed for real-time streaming video classification. Hosted as a Hugging Face Space, it processes live video input from a webcam, identifies and categorizes actions within the stream, and overlays the classification results directly onto the video feed. This application leverages the V-JEPA 2 model for its classification capabilities, offering a direct and interactive way to analyze video content. While the current live website indicates a runtime error preventing full functionality, its intended purpose is to provide immediate video action recognition, making it suitable for various real-time analysis scenarios.

ROAM1RealWorldAdversarialAttack is a Hugging Face Space developed by Artificio, designed to facilitate participation in competitions focused on real-world adversarial attacks. This application provides a centralized platform for users to access crucial competition details, explore dataset information, and track their performance on leaderboards. It also offers functionalities for managing submissions, ensuring a streamlined process for participants. Furthermore, users can review competition rules and update their team names directly within the application, making it a comprehensive tool for researchers and security professionals involved in assessing the robustness and vulnerabilities of AI systems through adversarial attack simulations.

Tiny Differentiable Simulator is a header-only C++ and CUDA physics library designed for reinforcement learning and robotics applications. It boasts zero dependencies, making it a lightweight and efficient solution for developers. The library implements various rigid-body dynamics algorithms, including forward and inverse dynamics, alongside contact models based on impulse-level LCP and force-based nonlinear spring-dampers. It also includes actuator models for motors, servos, and Series-Elastic Actuator (SEA) dynamics. The entire codebase is templatized, supporting automatic differentiation scalar types like CppAD, Stan Math fvar, and ceres::Jet, as well as regular float/double precision and fixed-point integer math for cross-platform deterministic computation. It can run thousands of simulations in parallel on a single RTX 2080 CUDA GPU at 50 frames per second and offers OpenGL 3+ and MeshCat visualizers.

Compare Depth Models is a Hugging Face Space designed for evaluating and comparing different depth estimation models, with a particular focus on Depth Anything and its predecessors. This tool is valuable for AI researchers and computer vision engineers who need to assess the performance and accuracy of various depth models. While the live website currently shows a runtime error, the intention of the tool is to provide a visual comparison of depth outputs from different models, aiding in research and development within the computer vision domain. It serves as a practical demonstration and comparison platform for advanced depth estimation techniques.

PolaroidVL 1.0 Demo offers a hands-on experience with a compact vision-language AI model, allowing users to interact directly by uploading images and posing questions. This tool is designed for detailed analysis and provides answers based on the visual and textual input. It supports common image formats like JPG, PNG, and GIF, with a file size limit of up to 10MB. Hosted on Hugging Face Spaces, it serves as an accessible platform for individuals interested in experimenting with AI's capabilities in understanding and interpreting visual information combined with natural language queries. It is particularly useful for educational purposes and research experimentation in the field of AI.

Grounding DINO Demo is a cutting-edge open-vocabulary object detection application hosted on Hugging Face Spaces. Users can upload an image and provide a text prompt to identify and highlight specific objects within that image. The tool then generates a marked-up image, visually indicating the detected objects based on the provided text. This makes it a valuable resource for researchers, developers, and AI enthusiasts working on computer vision tasks, particularly those involving object recognition and detection without pre-trained categories. It's an accessible way to experiment with advanced AI models for image analysis.

Reflection Llama 3.3 70B is an AI tool designed to execute Python scripts provided by the user. It operates by allowing users to set the 'MY_SCRIPT_CONTENT' environment variable with their desired Python script. The application then runs this script and displays the output. While the current live website indicates a runtime error and that the application does not appear to be initialized, the core functionality described suggests a tool for developers or technical users who need to run custom Python code within an AI environment. This could be useful for testing AI models, automating tasks, or performing data processing.

This article from SciTechDaily, titled "Can SpaceX Help NASA Reach Uranus Before It’s Too Late?", delves into how SpaceX's Starship could revolutionize a long-awaited mission to Uranus. It highlights Starship's significant advantages, including its heavy lift capacity, the ability to refuel in orbit, and its potential role as an aerobraking shield. The piece explains how these capabilities could drastically reduce travel time to Uranus, potentially cutting it in half to six and a half years, and eliminate the need for gravitational assists. The article also touches upon the scientific importance of exploring Uranus, its current unexplored status, and the challenges of funding and timing for such a mission, drawing on a study presented at the IEEE Aerospace Conference.

RPC is a Hugging Face Space designed for evaluating math problems with different AI reasoning models. This application allows users to select a dataset, a specific model, and other parameters to load and evaluate mathematical problems. The primary purpose is to demonstrate and experiment with AI models for a NeurIPS 2025 paper. Users can observe the performance and results of various reasoning approaches, making it a valuable tool for academic research and model development in the field of AI and mathematics. The platform provides a hands-on environment for researchers and students to interact with cutting-edge AI models.

ReinforcementLearning.jl is a comprehensive open-source package designed for reinforcement learning research within the Julia programming language. It emphasizes reusability and extensibility, offering elaborately designed components and interfaces that simplify the implementation of new algorithms. The package also facilitates easy experimentation, allowing users to run benchmark experiments, compare different algorithms, and evaluate agents efficiently. A core focus is on reproducibility, supporting a range of methods from traditional tabular approaches to modern deep reinforcement learning algorithms. It integrates several sub-packages like ReinforcementLearningBase.jl, ReinforcementLearningEnvironments.jl, and ReinforcementLearningCore.jl to provide a robust and modular framework for researchers and developers.