Projects

Overview: XScene-UEPlugin is a Unreal Engine 5 plugin designed for real-time visualization, management, editing, and scalable hybrid rendering of Gaussian Splatting models. This plugin aims to provide a comprehensive solution for reconstructing 3D scenes from multi-view photos. Key features: - Real-time Gaussian Splatting rendering using Niagara - Drag-and-drop .gspl assets into UE5 Content Browser - Fully Blueprint-compatible and extendable - Dynamic lighting and LOD generation - Hybrid rendering with native UE assets - Crop regions & pure VFX Niagara emitters Tech stack: Unreal Engine 5 Why it matters: This plugin offers a powerful tool for visual effects artists and developers to quickly and efficiently manage, edit, and render complex 3D scenes from multi-view photos, enhancing the realism and scalability of their projects. Repository: xverse-engine/XScene-UEPlugin GitHub stars: 1130

### Overview: This repository contains a collection of image-to-image papers, focusing on supervised and unsupervised methods. Papers are ordered by the arXiv first version submission time (if applicable). The repository is constantly updated, and contributors are encouraged to send pull requests or issues. ### Key features: - **Supervised**: - Pix2pix - TextureGAN - Contextual GAN - Pix2pix-HD - BicycleGAN - G2-GAN - Cross-domain disentanglement networks - GestureGAN - **Unsupervised**: - Unsupervised - General - Unsupervised - Attention/Instance guided - Unsupervised - Many-to-many (Attributes) - Unsupervised - Disentangled (and/or Exemplar guided) - Unsupervised - Video-to-Video Synthesis ### Tech stack: PyTorch, TensorFlow, and other relevant libraries. ### Why it matters: This collection provides a comprehensive overview of image-to-image techniques, useful for researchers and developers looking to understand and implement various methods for image-to-image translation and generation.

Overview: WebLLM Chat is a private AI chat interface that combines WebLLM with a user-friendly design, leveraging WebGPU to run large language models natively in your browser. It offers cutting-edge AI capabilities, guaranteed privacy, offline accessibility, and a feature-rich interface. Key features: - Browser-Native AI: Runs language models natively within your web browser with WebGPU acceleration. - GANranteed Privacy: AI model runs locally on your hardware, ensuring data never leaves your computer. - Offline Accessibility: Runs entirely offline after initial setup, allowing offline engagement. - Vision Model Support: Chat with AI by uploading and sending images. - User-Friendly Interface: Intuitive design with markdown support, dark mode, and responsive design. - Custom Models: Connects to any custom language model through MLC-LLM. - Open Source and Customizable: Built with NextChat, contributing to democratizing AI technology. Tech stack: TypeScript, WebGPU, Next.js, MLC-LLM. Why it matters: WebLLM Chat provides a private, accessible, and user-friendly AI chat experience, making it ideal for individuals and organizations looking to harness the power of AI without the need for servers or cloud services.

### Overview: MOVA is a foundation model designed to generate synchronized video and audio in a single model, offering a seamless and efficient solution for video and audio generation. ### Key features: - **Native Bimodal Generation**: MOVA generates high-fidelity video and synchronized audio in a single inference pass, eliminating error accumulation. - **Precise Lip-Sync & Sound FX**: Achieves state-of-the-art performance in multilingual lip-synchronization and environment-aware sound effects. - **Fully Open-Source**: The repository provides model weights, inference code, training pipelines, and LoRA fine-tuning scripts, making it accessible and reproducible. ### Tech stack: Python ### Why it matters: In a field dominated by closed-source models, MOVA is a significant contribution by making its model weights, inference code, and training pipelines available for open-source use, fostering innovation and collaboration in the field of video and audio generation.

### Overview: This repository is a curated list of research papers in visual grounding, focusing on the tasks of visual grounding and natural language object retrieval. It aims to provide a comprehensive overview of the field and ensure the quality of the papers by reviewing them before adding them to the repository. ### Key Features: - **Papers**: Over 100 papers on visual grounding and natural language object retrieval. - **Code**: Links to the code for some papers. - **Chronological Order**: Papers are organized in chronological order to provide a clear roadmap. - **Contributing**: A detailed guide on how to contribute papers, including fork, change README, and submit pull requests. ### Tech Stack: - **Language**: Unknown - **Stars**: 1128 ### Why it Matters: This repository serves as a valuable resource for researchers and practitioners in the field of visual grounding, providing a curated list of relevant papers and a clear roadmap for the field.

Overview: DataDreamer is a Python library designed for generating synthetic data, training models, and aligning them with prompts. It simplifies complex workflows, making it accessible and research-grade. Key features: - Synthetic data generation - Prompt-based data creation - Efficient training and model alignment Tech stack: Python Why it matters: DataDreamer streamlines the process of generating synthetic data and training models, making it easier for developers to experiment with NLP tasks and contribute to research projects.

Overview: PhotonCamera is an Android Camera application that utilizes advanced image processing techniques for improved image quality and performance. Key features: - Enhanced image processing - Improved camera functionality - Customizable user interface Tech stack: Java Why it matters: PhotonCamera offers a significant enhancement in camera functionality and image processing capabilities, making it a valuable tool for developers looking to improve their Android app's performance and user experience.

Overview: Cutie is a video object segmentation framework that builds upon XMem, offering improved consistency, robustness, and speed. The repository includes code for standard video object segmentation and a GUI tool for interactive video segmentation. Key features: - Improved segmentation consistency - Enhanced robustness - Faster processing speed - GUI tool with permanent memory option for better control Tech stack: Python Why it matters: Cutie provides a powerful tool for video object segmentation, making it easier and faster to analyze video content by identifying and marking objects.

Overview: This repository contains 30 runnable Jupyter notebooks that cover various memory techniques for Large Language Models (LLMs), including conversation buffers, vector stores, knowledge graphs, episodic and semantic memory, working memory, MemGPT, Mem0, Letta, Zep, Graphiti, LoCoMo benchmarks, and production patterns. Key features: - 30 Jupyter notebooks covering multiple memory techniques - Detailed explanations and runnable code examples - Comprehensive coverage of agent memory for LLMs Tech stack: Jupyter Notebook Why it matters: This repository is essential for developers and researchers looking to enhance their understanding of how to build robust and efficient memory systems for LLMs, providing practical, hands-on learning opportunities for both beginners and advanced users.

### Overview: Atlassian MCP Server is a secure, cloud-hosted bridge that connects Atlassian's various products like Jira, Confluence, and Bitbucket to AI tools such as Claude, ChatGPT, Cursor, VS Code, and more. This official server ensures seamless, real-time access to your AI tools using OAuth 2.1 or API tokens. ### Key Features: - **Secure Access**: Provides secure, real-time access to Atlassian's products and AI tools. - **Model Context Protocol (MCP)**: Compatible with the Model Context Protocol, enabling seamless integration. - **OAuth 2.1 or API Tokens**: Supports multiple authentication methods for secure connections. ### Tech Stack: JavaScript ### Why It Matters: Atlassian MCP Server is essential for businesses looking to integrate their Atlassian products with AI tools, ensuring a secure and efficient workflow. It simplifies the integration process and provides a reliable, cloud-based solution for remote collaboration and productivity.

Overview: This project offers production-ready AI skills and tools for creating images, videos, audio, and 3D assets, including support for popular software like Blender, Maya, ZBrush, Unreal Engine, and Unity. The skills are designed to ensure consistency and quality across different projects. Key features: - Scenario MCP server integration - Expert teams building and verifying skills - Support for DCC software and game engines - Comprehensive documentation and support Tech stack: Python Why it matters: This project streamlines the process of creating high-quality content using AI, ensuring consistency and reliability across various projects and software tools. Repository: scenario-labs/skills Primary language: Python GitHub stars: 1088 GitHub description: Get production-ready images, video, audio, and 3D from any AI agent: skills that pick the right model, price before spending, and keep characters and brands consistent through the Scenario MCP, plus expert teams that drive Blender, Maya, ZBrush, Unreal, and Unity.

Overview: This project is an AI-powered knowledge management system designed to help users maintain and organize their information efficiently. It includes guides, starter vaults, agent skills, and scripts for a self-organizing knowledge base. Key features: - A guide for setting up a knowledge base that an AI agent maintains for you. - A guide for creating and managing agents. - A guide for understanding and using handbooks for advanced techniques and tools. Tech stack: HTML Why it matters: This project is a comprehensive guide for setting up and maintaining an AI-powered knowledge management system, making it easier for users to organize and access their information efficiently. Repository: undefined-ui/second-brain-os GitHub stars: 1110 GitHub description: An AI second brain that maintains itself. Full guide, starter vault, agent skills and scripts for a self-organizing knowledge base in Claude Code and Obsidian.

### Overview: AI Waifu Vtuber & Assistant is an AI virtual YouTuber that can respond to your voice and chat with Twitch viewers. It supports various languages and can be integrated with VoiceVox, DeepL, Whisper OpenAI, Seliro TTS, and VtubeStudio. ### Key Features: - **Voice Response**: AI Waifu can respond to your voice commands. - **Chat Integration**: It can read and answer to Twitch chat. - **Language Support**: Supports Japanese, Russian, English, German, Spanish, French, Tatar, Ukrainian, Uzbek, Kalmyk, and Indic languages using Seliro TTS. - **TTS Engines**: Compatible with VoiceVox, DeepL, Whisper OpenAI, and Seliro TTS. ### Tech Stack: - Python ### Why It Matters: AI Waifu Vtuber & Assistant is a powerful tool for creating engaging virtual YouTubers and streamers. It can enhance your presence on social media platforms and provide a personalized, conversational experience for your audience.

Overview: This tutorial is designed for coders interested in understanding the building blocks of large language models (LLMs), their implementation, and how to code them from the ground up using PyTorch. The workshop covers an introduction to LLMs, their use cases, and provides hands-on coding of a small GPT-like LLM, including its data input pipeline, core architecture components, and pretraining code. Key features: - Introduction to LLMs and their use cases - Coding a text input pipeline and a custom PyTorch DataLoader - Assembly of LLM building blocks into a GPT-like model - Pretraining of the model architecture - Loading pretrained weights into the self-implemented architecture - Introduction to LitGPT library for finetuning LLMs Tech stack: PyTorch, Jupyter Notebook Why it matters: This workshop offers a practical, step-by-step guide to understanding and implementing LLMs, making it valuable for developers looking to build their own models or improve their understanding of existing models.

Overview: AI-Trader is a powerful backtesting framework for algorithmic trading, built on Backtrader, featuring 20+ strategies, multi-market support, and an integrated MCP server for professional traders. Key features: - Config-Driven Workflows: Define and manage backtests with version-controllable YAML files for reproducible results. - Seamless LLM Integration: Built-in MCP (Model Context Protocol) server allows AI assistants like Claude to run backtests, fetch data, and analyze strategies. - Multi-Market Support: Test strategies on US stocks, Taiwan stocks, cryptocurrencies, and forex. - Extensive Strategy Library: Comes with over 20 built-in strategies, from classic indicators to advanced adaptive models. - Powerful CLI: A rich command-line interface to run backtests, fetch market data, and list strategies. - Developer Friendly: Easily create and test custom strategies with simple helpers and a clear structure. Tech stack: Python Why it matters: AI-Trader provides a professional, config-driven backtesting framework that allows traders to easily test, optimize, and integrate trading strategies with Large Language Models across various markets, including stocks, cryptocurrencies, and forex. Repository: whchien/ai-trader GitHub stars: 1116

### Overview: DL-Hub 是一个专注于机器学习、深度学习、计算机视觉、自然语言处理和大模型项目实战的仓库。它包含50篇关于这些领域的文章，旨在帮助开发者循序渐进地学习和实践。 ### Key features: - 791 注册 ID 的图像分类、检测和分割模型 - 814 注册 ID 的分类、命名实体识别和阅读理解模型 - 64 注册 ID 的图分类、节点嵌入和推荐模型 - 50+ 论文笔记的因果语言模型、指令调优、前缀调优和Mamba模型 ### Tech stack: Python ### Why it matters: DL-Hub 是一个全面且实用的资源库，涵盖了从基础到高级的深度学习项目，帮助开发者在学习和实践中快速上手。

Overview: The WIT dataset is a large, multimodal multilingual dataset comprising 37.6 million image-text examples with 11.5 million unique images across 108 Wikipedia languages. It is the largest publicly available multimodal dataset by the number of image-text examples at the time of this writing. Key features: - 37.6 million image-text examples - 11.5 million unique images across 108 languages - Curated and annotated for quality - Includes page-level metadata and contextual information Tech stack: Python, PyTorch, TensorFlow, and other standard machine learning tools. Why it matters: The WIT dataset is crucial for researchers and developers looking to train and evaluate multimodal models, as it provides a comprehensive and diverse set of image-text examples that can be used for pretraining and downstream tasks.

### Overview: Visual Pushing and Grasping (VPG) is a method for training robotic agents to learn how to plan complementary pushing and grasping actions for manipulation. This approach leverages deep reinforcement learning to enable robots to make efficient and precise manipulations in real-world scenarios. ### Key Features: - **Self-Supervised Learning**: VPG operates directly on visual observations, learning from trial and error without explicit labels. - **Fully Convolutional Networks**: Two networks map visual observations to actions, one for pushing and one for grasping. - **Joint Training**: Both networks are trained jointly in a Q-learning framework, using self-supervised rewards from successful grasps. - **Real-World Applications**: The repository provides PyTorch code for training and testing VPG policies on a UR5 robot arm in simulation and real-world settings. ### Tech Stack: Python ### Why It Matters: This method demonstrates the potential for robots to learn complex synergies between non-prehensile and prehensile actions through self-supervised deep reinforcement learning, making manipulation more efficient and versatile.

Overview: This project is a deep learning pipeline designed to classify the genre of songs based on your music library. It leverages TensorFlow and Tensorflow-Learn to create a dataset from your mp3 files and train a model to predict genres. Key features: - Dataset creation from your own music library - Model training and testing - Pipeline for labeling new songs with the model Tech stack: Python Why it matters: This project is useful for music enthusiasts and artists who want to understand the genre of songs they listen to, or for those interested in using deep learning for music classification. Repository: despoisj/DeepAudioClassification Primary language: Python GitHub stars: 1114

### Overview: A real-time monitoring dashboard for Claude Code, Cursor, and Codex, providing insights into sessions, agent activity, tool usage, and more. ### Key features: - Real-time monitoring of Claude Code, Cursor, and Codex sessions. - Detailed tracking of agent hierarchies, token usage, and orchestration. - Interactive web UI and native app for macOS and Windows. - Live analytics, Kanban status board, notifications, and a cute buddy. - Tracks tool usage, session duration, and cost. - Supports Node.js, TypeScript, and SQLite3. ### Tech stack: TypeScript, Node.js, Express, React, Vite, TailwindCSS, WebSockets, SQLite3. ### Why it matters: This dashboard offers a comprehensive solution for developers tracking and managing AI coding agents, providing real-time insights and easy-to-navigate monitoring tools.