Projects

### Overview: Claude Skills Marketplace is a curated marketplace of Claude Code plugins designed to streamline software engineering workflows. It offers a variety of plugins for Git automation, testing, and code review, making it an essential resource for developers. ### Key Features: - **Git Automation**: Automates common Git tasks such as pushing, pulling, and merging. - **Code Review**: Facilitates code review processes with detailed feedback and suggestions. - **Testing**: Enhances testing workflows with automated tests and integration with CI/CD pipelines. ### Tech Stack: The repository uses the following technology stack: - **Claude**: The Claude Code plugin marketplace. - **GitHub**: The hosting platform for the repository. ### Why It Matters: Claude Skills Marketplace provides a comprehensive set of tools to enhance software engineering workflows, making it easier and more efficient to manage projects, automate tasks, and improve code quality.

### Overview: Agent Skills revolutionize AI agent architecture by providing modular, standardized capabilities. ### Key Features: - **Progressive Disclosure**: Lightweight metadata loads early, full instructions load only when relevant, and supporting resources are accessed when needed. - **Skill-Based Architecture**: Agent Skills are modular, standardized `SKILL.md` packages that give agents on-demand capabilities via progressive disclosure. - **Context Management**: Skills can be used in various environments, including Claude, Anthropic, and other AI platforms. ### Tech Stack: `agent-skills`, `ai-agents`, `skill-md`, `progressive-disclosure`, `context-management`. ### Why it Matters: Agent Skills simplify the development and deployment of AI agents by providing a consistent and scalable way to integrate modular capabilities. This approach ensures that agents can be quickly and efficiently configured to meet specific tasks, without the need for extensive setup or customization.

Overview: iOS MCP 是一个运行在越狱 iPhone 上的 MCP 服务器，支持多种自动化操作，包括触控手势、硬件按键、文字输入、截图、应用管理、无障碍操作、剪贴板管理、文件系统操作、日志读取、设备控制和设备信息查询等，适用于自动化测试和iOS逆向调试。 Key features: - 精确屏幕坐标操作 - 物理按键模拟 - 剪贴板快速输入 - 屏幕截图与OCR识别 - 启动/关闭应用 - 查询App沙盒/容器路径与entitlements - 获取UI元素树与定位元素 - 识别与描述屏幕内容 - 读写剪贴板 - 目录列举与文件读写 - 获取系统日志与崩溃日志 - 亮度、音量控制 - 设备信息查询 - 打开URL与Shell命令执行 Tech stack: Objective-C Why it matters: iOS MCP 是一个强大的工具，能够简化和加速iOS设备的自动化测试和逆向调试过程，提高了开发效率和代码质量。它提供了全面的功能，使得开发者能够轻松地控制和操作iOS设备，从而进行更深入的代码分析和功能测试。

Overview: DL3DV-10K is a large-scale dataset of real-world videos featuring 51.2 million frames from 10,510 videos captured from 65 types of point-of-interest (POI) locations, covering both bounded and unbounded scenes, with different levels of reflection, transparency, and lighting. It serves as a comprehensive benchmark for recent NVS methods. Key features: - 10,510 multi-view scenes covering 51.2 million frames at 4k resolution. - 140 videos as Novel view synthesis (NVS) benchmark. - All videos are annotated by scene environment (indoor vs. outdoor), levels of reflection, transparency, and lighting. - Released samples include colmap calculated camera pose. - Benchmark videos offer trained parameters from the SOTA NVS methods, including 3D Gaussian Splatting, ZipNeRF, Mip-NeRF 360, Instant-NGP, and Nerfacto. Tech stack: HTML Why it matters: DL3DV-10K provides a comprehensive benchmark for recent NVS methods, enabling researchers to evaluate and improve their models.

Overview: This repository contains a real-time 3D multi-person pose estimation demo written in PyTorch. It utilizes the Intel OpenVINO backend for fast inference on CPUs, making it suitable for applications requiring low latency and high performance. Key features: - Real-time 3D multi-person pose estimation - Supports detection of up to 18 types of keypoints - Achieves 100 mm MPJPE on the CMU Panoptic subset - Available pre-trained model on Google Drive Tech stack: PyTorch, OpenVINO, OpenCV, CMake, Python, C++ Compiler Why it matters: This demo is highly valuable for applications needing accurate 3D human pose estimation, especially in scenarios where real-time performance is crucial, such as in augmented reality, robotics, or virtual reality environments. Repository: Daniil-Osokin/lightweight-human-pose-estimation-3d-demo.pytorch GitHub stars: 688 GitHub description: Contains a 3D multi-person pose estimation demo in PyTorch, utilizing OpenVINO for fast inference on CPUs. The demo is based on the Lightweight OpenPose and Single-Shot Multi-Person 3D Pose Estimation From Monocular RGB papers, achieving 100 mm MPJPE on the CMU Panoptic subset. The major part of this work was done by Mariia Ageeva when she was an intern at Intel. Requirements: Python 3.5 or above, CMake 3.10 or above, C++ Compiler, OpenCV

Overview: 一个生产级、高性能、模块化、可扩展的中文NLP工具包，包括中文分词、平均感知机、fastText、拼音、新词发现、分词纠错、BM25、人名识别、命名实体、自定义词典等功能。 Key features: - 中文分词 - 平均感知机 - fastText - 拼音 - 新词发现 - 分词纠错 - BM25 - 人名识别 - 命名实体 - 自定义词典 Tech stack: Java Why it matters: 该工具包是生产级、高性能、模块化、可扩展的中文NLP解决方案，适用于各种应用场景，如文本处理、信息检索和自然语言理解等。

Overview: Overmind is a platform that continuously trains and improves AI agents using data from production traces and other datasets. It provides a unified API for fine-tuning models, benchmarking them against evaluation metrics, and serving them via a 1-pan platform. Key features: - Continuous training and improvement of AI agents - Integration with production traces and other datasets - Unified API for fine-tuning models - Benchmarking against evaluation metrics - Serving through a 1-pan platform Tech stack: Python Why it matters: Overmind is designed to help developers continuously improve their AI agents by leveraging data from production environments. It simplifies the process of fine-tuning models, making it easier to create and deploy high-quality AI solutions.

Overview: This repository contains shared assets for the Ultralytics YOLO ecosystem, including logos, media, sample images, pretrained model weights, dataset artifacts, and release assets. Key features: - Visual assets - Model assets - Dataset assets Tech stack: GitHub Actions workflow for formatting and documentation. Why it matters: This repository provides stable files for the Ultralytics YOLO ecosystem, ensuring consistency and ease of use across various documentation, applications, and releases.

### Overview: AttentionGAN: Unpaired Image-to-Image Translation using Attention-Guided Generative Adversarial Networks ### Key Features: - Unpaired Image-to-Image Translation - Multi-Domain Image-to-Image Translation - Uses Attention-Guided Generative Adversarial Networks (GANs) - State-of-the-Art Performance in Selfie to Anime, Horse to Zebra, Zebra to Horse, Apple to Orange, Orange to Apple, Map to Aerial Photo, Aerial Photo to Map, and Style Transfer ### Tech Stack: Python ### Why it Matters: AttentionGAN offers a novel approach to image-to-image translation by incorporating attention mechanisms, significantly improving the quality and diversity of generated images across various domains and scenarios.

Overview: SkyChat是一款基于中文GPT-3 API的聊天机器人项目，支持中文-英文互译、内容续写、对对联、写古诗等多种功能。 Key features: - 中文-英文互译 - 内容续写 - 对对联 - 写古诗 Tech stack: C# Why it matters: SkyChat为开发者提供了一个强大的中文聊天机器人解决方案，适用于多种应用场景，如内容续写、对联创作等，极大地提升了工作效率和用户体验。

### Overview: Voice Builder is an open-source text-to-speech (TTS) voice building tool designed for simplicity, flexibility, and collaboration. It enables anyone with basic computer skills to experiment with voice training and listen to the resulting synthesized voice, aiming to reduce barriers in creating new voices and accelerating TTS research. ### Key Features: - **Text-to-Speech (TTS) Voice Building Tool** - **Simplicity and Flexibility** - **Collaborative Experimentation** - **User-Friendly Interface** - **Custom Data Exporter Option** ### Tech Stack: - JavaScript ### Why It Matters: Voice Builder simplifies the process of creating and experimenting with TTS voices, making it easier for researchers and developers to collaborate and accelerate their work in the field of text-to-speech technology. ### GitHub Stars: 687 ### Repository: - **Repository:** google/voice-builder - **Primary Language:** JavaScript - **GitHub Stars:** 687 ### Additional Information: - **Installation:** - Prerequisites: Create a project on Google Cloud Platform (GCP), enable billing, install Docker, Firebase CLI, Node.js, and Google Cloud SDK. - Deployment: Clone the project, log in to Google Cloud and Firebase, create GCS buckets, deploy cloud functions and UI components. - Create an Example Voice: Follow the steps to set up and access your Voice Builder instance via the provided IP address. - **Additional Information:** Optional: Set up a custom data exporter for more detailed experimentation. - **Publication:** Voice Builder was

### Overview: This repository is a curated list of AI tools, agent skills, libraries, and resources specifically designed for economics research, teaching, and policy analysis. Maintained by the OpenEcon team, it provides a comprehensive collection of projects and skills that are valuable for economists. ### Key features: - **AI-research-setup**: A step-by-step guide to setting up Claude Code and Codex for economics research on macOS and Windows. - **OpenEcon Data**: A MCP server and web app for plain-English queries over 330,000+ indicators from FRED, World Bank, IMF, Eurostat, and more. - **Stata-MCP**: Runs Stata from VS Code, Cursor, Claude Code, Codex, and Copilot with real-time output, a data viewer, and graph display. - **Econ Writing Skill**: A synthesizing guide for economics paper-writing skills, including 50+ guides by Cochrane, McCloskey, Shapiro, Head, and others. - **Econ Paper Review Skill**: Writes referee reports on economics papers with verified comments, editing notes, and a step-by-step revision plan; free for non-commercial use. - **Econ Slides Skill**: Turns a paper into a Beamer talk with a timed speaker script, including job-market and discussant decks. - **overleaf-sync-now**: An agent skill and CLI that pulls recent Overleaf web edits into a Dropbox-synced copy before AI agents edit it. - **Econ Auto Research**: A vision-stage plan (no code

### Overview: This repository is a curated list of papers on adaptation strategies for agentic AI systems, specifically focusing on the "Adaptation of Agentic AI" paper. ### Key features: - **Agent Adaptation:** - **A1: Tool Execution Signaled** - Development Timeline: 2025.11 - Methods: RL-based - Venue: arXiv - Task(s): IR - Tool(s): Retrievers - Agent Backbone: LFM2 - Tuning: GRPO - **A2: Agent Output Signaled** - Development Timeline: 2025.10 - Methods: RL-based - Venue: arXiv - Task(s): Document OCR - Tool(s): Synthetic Document Verifier - Agent Backbone: Qwen2.5-VL - Tuning: SFT, GRPO - **T1: Agent-Agnostic Tool Adaptation** - Development Timeline: 2025.10 - Methods: RL-based - Venue: Nature’25 - Task(s): Formal Theorem Proving - Tool(s): Lean Compiler - Agent Backbone: Transformer (3B Enc-Dec) - Tuning: SFT, AlphaZero, TTRL - **T2: Agent-Supervised Tool Adaptation** - Development Timeline: 2025.10 -

### Overview: BundleTrack is an innovative 6D pose tracking framework designed for novel objects without the need for instance or category-level 3D models. This project addresses a significant challenge in pose tracking by leveraging recent advancements in deep learning for segmentation and robust feature extraction, coupled with memory-augmented pose graph optimization. The approach enables long-term, low-drift tracking under various challenging scenarios, including significant occlusions and object motions. Comprehensive experiments on two public benchmarks demonstrate that BundleTrack significantly outperforms state-of-the-art methods, achieving comparable performance to category-level 6D tracking or dynamic SLAM methods while requiring reduced information requirements. ### Key Features: - **General Framework for Novel Object Pose Tracking**: BundleTrack is a versatile framework that can track the pose of novel objects without the need for pre-existing 3D models. - **Complementary Advancements**: It leverages recent advances in deep learning for segmentation and robust feature extraction. - **Memory-Augmented Pose Graph Optimization**: This technique ensures spatiotemporal consistency, enabling long-term tracking under challenging conditions. ### Tech Stack: - **Programming Language**: C++ - **GitHub Stars**: 687 ### Why It Matters: BundleTrack represents a significant advancement in pose tracking by addressing the limitations of traditional methods that require pre-existing 3D models. By providing a solution that can track novel objects without such constraints, it opens up new possibilities for applications in robotics, augmented reality, and more. The framework's ability to perform well under various challenging scenarios, including significant occlusions and object motions, makes it highly applicable

Overview: GlueStick is an innovative joint deep matcher designed for robust image matching, specifically for points and lines. It was presented at ICCV 2023 and is now available in a new training framework, GlueFactory. Key features: - Joint deep matcher for points and lines - Robust image matching - Available in Jupyter Notebook implementation - Supports multiple feature extractors, line extractors, robust estimators, and evaluations on benchmarks Tech stack: Jupyter Notebook Why it matters: GlueStick offers a novel approach to image matching, particularly effective for points and lines, and is now accessible through GlueFactory, making it easier to train and evaluate.

**Overview:** OmniVinci is a project aimed at enhancing the architecture and data of large language models (LLMs) to better understand and interact with the world through multiple modalities such as vision, audio, and language. **Key Features:** - OmniAlignNet for aligning vision and audio embeddings in a shared latent space. - Temporal Embedding Grouping for capturing temporal relationships between vision and audio signals. - Constrained Rotary Time Embedding for encoding absolute temporal information in omni-modal embeddings. **Tech Stack:** Python **Why It Matters:** OmniVinci aims to bridge the gap between vision, audio, and language by creating a unified model that can understand and interact with the world in a more holistic manner, which is crucial for advancing machine intelligence and enabling more effective communication and understanding across different modalities.

Overview: Structured3D is a large-scale dataset containing 3.5K house designs, including photo-realistic 2D images and ground truth annotations for semantic, albedo, depth, surface normal, and layout. This dataset is designed for structured 3D modeling research. Key features: - Large-scale photo-realistic dataset - 3.5K house designs - Ground truth annotations for semantic, albedo, depth, surface normal, and layout Tech stack: Python Why it matters: Structured3D provides a comprehensive and high-quality dataset for researchers working on structured 3D modeling, enabling them to evaluate and improve their models through real-world data.

### Overview: This project is a comprehensive repository for Skeleton-based Action Recognition, featuring a wide range of papers, datasets, and tools. It serves as a valuable resource for researchers and developers interested in the field of action recognition using skeleton-based methods. ### Key Features: - **Datasets:** Includes popular datasets like NTU RGB+D 120, NTU60-X, PoseC3D 2D Skeleton Dataset, and others. - **Papers:** A collection of papers on supervised, semi-supervised, and unsupervised methods for skeleton-based action recognition. - **Leaderboards:** Provides performance metrics for various datasets, including NTU-RGB+D and NTU-RGB+D 120. - **Repositories:** Offers resources and models for top-performing papers, including pre-trained models and interactive dashboards. ### Tech Stack: HTML ### Why it Matters: This repository serves as an essential resource for researchers and developers looking to advance the field of action recognition using skeleton-based methods. It provides a wealth of information, datasets, and tools that can significantly contribute to the development and improvement of this technology. ### GitHub Stars: 685 ### Primary Language: HTML ### Repository: niais/Awesome-Skeleton-based-Action-Recognition ### TODO: - Paper list - Supervised methods - Semi-supervised methods - Unsupervised methods - Adversarial methods - Leaderboard for supervised methods - NTU RGB+D - NTU RGB

Overview: Rankify is a comprehensive Python toolkit designed for unified retrieval, re-ranking, and retrieval-augmented generation (RAG) research. It integrates 40 pre-retrieved benchmark datasets and supports 7 retrieval techniques, 24 state-of-the-art re-ranking models, and multiple RAG methods. Key features: - Integrates 40 pre-retrieved benchmark datasets - Supports 7 retrieval techniques - 24 state-of-the-art re-ranking models - Multiple RAG methods - Flexible generator architecture for seamless experimentation and benchmarking Tech stack: Python Why it matters: Rankify provides a modular and efficient framework for researchers and practitioners in the field of retrieval, re-ranking, and RAG tasks. It enables seamless experimentation and benchmarking across different models and methods, making it a powerful resource for researchers and practitioners in the field. GitHub stars: 684

Overview: This repository contains a comprehensive collection of papers from the INTERSPEECH 2023-24 conference, focusing on speech and language processing research. It includes the latest advancements and includes code for further exploration. Key features: - Papers from INTERSPEECH 2023-24 - Latest research in speech and language processing - Code included for deeper understanding Tech stack: Python, Jupyter Notebook, and other relevant tools for data analysis and research. Why it matters: This repository serves as a valuable resource for researchers, students, and enthusiasts interested in the latest developments in speech and language processing. It supports the advancement of speech technology through the sharing of cutting-edge research and practical applications.