Projects

### Overview: MeshAnything is an innovative project that leverages autoregressive transformers to generate artist-created meshes. This breakthrough work was presented at ICLR 2025 and offers a new perspective on mesh generation, particularly for applications that require high-quality and unique designs. ### Key Features: - **Artist-Created Mesh Generation**: The project allows users to create and generate meshes that closely resemble human artistic styles. - **Autoregressive Transformers**: Utilizes advanced neural architectures to enhance the quality and diversity of generated meshes. - **Versatile Input Formats**: Supports various input formats, including OBJ, PLY, and others, making it highly adaptable for different applications. ### Tech Stack: The project is primarily implemented using Python, with the following dependencies: - PyTorch - Flash-Attn - Gradio for interactive demos ### Why It Matters: MeshAnything offers a significant advancement in the field of mesh generation, providing artists and designers with a powerful tool to create unique and high-quality meshes directly from their artistic styles. This technology has the potential to revolutionize applications in fields such as game development, architecture, and more. ### Repository: - **Repository**: `buaacyw/MeshAnything` - **Stars**: 2307 - **Description**: "ICLR 2025] From anything to mesh like human artists. Official impl. of "MeshAnything: Artist-Created Mesh Generation with Autoregressive Transformers" ### Installation: To set up your environment, follow these steps: ```bash git clone https://github.com/buaacy

### Overview: CLAIMED is a low-code/no-code rapid prototyping framework designed to seamlessly integrate and deploy AI components into production environments. It automates the creation of container images, installation of dependencies, and orchestration of workflows using Kubernetes and OpenShift. ### Key Features: - **Code Transformation**: Takes Jupyter notebooks, Python scripts, and R scripts as input and automatically converts them into executable AI components. - **Container Image Creation**: Automatically generates container images and pushes them to container registries. - **Dependency Management**: Ensures all required dependencies are installed within the container images. - **Pipeline Creation**: Creates KubeFlow Pipeline components, targeting various workflow execution engines. - **Job Configuration**: Generates Kubernetes job configurations for execution on Kubernetes/Openshift clusters. - **CI/CD Integration**: Seamlessly integrates with CI/CD pipelines to trigger component creation and deployment. ### Tech Stack: CLAIMED utilizes Jupyter Notebook for development and deployment. ### Why It Matters: CLAIMED simplifies the process of creating and deploying AI components, making it accessible to developers without extensive coding knowledge. Its integration with CI/CD pipelines ensures that components are consistently updated and deployed in production environments. ### Repository: Repository: claimed-framework/claimed Primary Language: Jupyter Notebook GitHub Stars: 2306 ### Getting Started: - **Install**: `pip install claimed` - **Usage**: Run `c3_create_operator "<your-operator-script>.py" --repository "<registry>/<namespace>"` for your Python script or notebook. - **Help**: `c3_create_operator --help`

### Overview: TorchCAM is a comprehensive toolset for generating class activation maps (CAMs) and token activation maps for various PyTorch models, including Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs). It provides 12 CAM methods, including Grad-CAM, Grad-CAM++, Smooth Grad-CAM++, Score-CAM, SS-CAM, IS-CAM, XGrad-CAM, Layer-CAM, Finer-CAM, LeGrad, and RefineCAM, as well as token activation maps for Vision Language Models (VLMs). ### Key Features: - **Class Activation Maps (CAMs)**: Generate visual explanations for model predictions. - **Token Activation Maps**: Visualize the contribution of each token in a language model. - **Automatic Target-Layer Resolution**: Reshape transforms for ViTs to ensure accurate activation maps. - **Integration with PyTorch**: Seamless integration with PyTorch models, including CNNs and ViTs. - **Versatile Applications**: Suitable for image classifiers, detectors, segmenters, and VLMs. ### Tech Stack: - Python ### Why It Matters: TorchCAM offers a powerful and versatile tool for understanding and visualizing the decision-making process of various PyTorch models, making it invaluable for researchers, developers, and practitioners in the field of computer vision and natural language processing. ### Repository: Repository: frgfm/torch-cam Primary Language: Python GitHub Stars: 2306 ### Documentation: - **Documentation Status**: [Docs Status](https

Overview: Lazynlp is a Python library designed to scrape and clean web pages, enabling the creation of massive monolingual datasets. It simplifies the process of obtaining URLs, deduplicating them, and downloading pages from various sources like Reddit, Gutenberg, and Wikipedia. Key features: - Crawling and cleaning web pages - Deduplicating URLs - Downloading web pages from various sources - Creating datasets larger than those used by OpenAI for GPT-2 Tech stack: Python Why it matters: This library streamlines the process of obtaining and cleaning data from the web, making it easier to build large datasets that can be used for various applications, including language modeling and text analysis. Repository: chiphuyen/lazynlp Primary language: Python GitHub stars: 2305

### Overview: The Fable Method is a set of skills and techniques distilled from Claude Fable 5's final days, designed to help models think, act, and prove their work. It includes four key skills: think (fable-method), act (fable-loop), prove (fable-judge), and grow (fable-domain). The method is evaluated against itself through fifteen rounds and over 260 agent runs, with results documented in the `eval` directory. ### Key Features: - **Think (fable-method)**: Models classify asks before acting. - **Act (fable-loop)**: Models define done with named verification. - **Prove (fable-judge)**: Models gather evidence in parallel and verify by observation. - **Grow (fable-domain)**: Models commit to one recommendation and change the smallest correct thing. ### Tech Stack: - Python ### Why It Matters: The Fable Method provides a structured approach for models to think, act, and prove their work, ensuring that the final output is reliable and verifiable. It helps models avoid common pitfalls and ensures that their decisions are based on evidence and context.

Overview: Dive into the fundamentals of machine learning, deep learning, and mathematical optimization through hands-on Python projects. Key features: - Build machine learning models from scratch - Understand concepts like gradient descent, logistic regression, and k-means clustering - Explore practical applications using Jupyter notebooks and exercises - Access lecture slides and supplementary resources Tech stack: Python Why it matters: Master the essential skills needed for building and analyzing machine learning models, backed by a comprehensive textbook and practical resources. Repository: neonwatty/machine-learning-refined GitHub stars: 2304 GitHub description: Master the fundamentals of machine learning, deep learning, and mathematical optimization by building key concepts and models from scratch using Python.

Overview: MarkPDFDown is a tool designed to convert PDF documents into clean, editable Markdown text, leveraging advanced multimodal AI models through LiteLLM to ensure accurate text extraction, formatting preservation, and handling of complex document structures. Key features: - PDF to Markdown Conversion - Image to Markdown Conversion - Multi-Provider Support - Flexible CLI - Format Preservation - Page Range Selection - Modular Architecture Tech stack: Python Why it matters: MarkPDFDown simplifies the process of converting PDF documents into Markdown format, making it easier for developers and content creators to work with clean, editable content directly from their PDFs.

### Overview: Paperlib is an open-source academic paper management tool designed to simplify the process of managing and citing conference papers, especially those without DOI or ISBN. It aims to provide a modern, user-friendly interface and robust metadata scraping capabilities to help researchers easily find and manage their citations. ### Key Features: - **Metadata Scraping**: Paperlib scrapes metadata from various sources like Google Scholar and DBLP, ensuring accurate citation information. - **Fulltext and Advanced Search**: Enables fulltext search and advanced filtering options for better research focus. - **Smart Filters**: Enhances search efficiency with smart filters tailored for different disciplines. - **Rating, Flag, Tagging, Folders, and Notes**: Organizes papers with features like rating, flagging, tagging, and note-taking. - **RSS Feed Subscription**: Allows following the latest publications in your research area. - **PDF Downloads**: Automatically locates and downloads PDF files from the web. - **MacOS Spotlight-like Plugin**: Facilitates easy reference management while writing papers. - **Cloud Sync**: Supports macOS, Linux, and Windows for seamless cloud synchronization. - **Beautiful UI**: Clean and modern interface with extensions for additional features. ### Tech Stack: TypeScript ### Why It Matters: Paperlib aims to address the challenges faced by researchers in managing and citing conference papers, especially those without standard identifiers. By providing a user-friendly interface and robust metadata scraping capabilities, it simplifies the citation process and helps researchers focus on their research without the hassle of manual searches. ### Repository: Future-Scholars/paperlib ### Primary Language:

Overview: This project is an official implementation of a paper that extracts triangular 3D models, materials, and lighting from images. It leverages the Kaolin library for GPU acceleration and supports differentiable marching tetrahedrons. Key features: - Joint optimization of topology, materials, and lighting - Adapted code from NVIDIA's Kaolin library - Supports FlexiCubes isosurfacing technique Tech stack: Python Why it matters: This project is significant for researchers and developers working on 3D modeling and image processing, as it provides a robust and efficient solution for extracting complex 3D models from images. Repository: NVlabs/nvdiffrec GitHub stars: 2301 GitHub description: Official code for the CVPR 2022 (oral) paper "Extracting Triangular 3D Models, Materials, and Lighting From Images". License: Copyright &copy; 2022, NVIDIA Corporation. All rights reserved. This work is made available under the Nvidia Source Code License.

Overview: Emgu CV is a cross-platform .NET wrapper for OpenCV, enabling developers to use OpenCV functions in C#. Key features: - Cross-platform compatibility - Compilable with Visual Studio, Unity, and "dotnet" command - Support for Windows, Mac OS, Linux, iOS, and Android Tech stack: C# Why it matters: Emgu CV simplifies the integration of OpenCV into .NET applications, making complex image processing tasks accessible to developers using C#. Repository: emgucv/emgucv GitHub stars: 2300

Overview: Fairlearn is a Python package designed to assess and improve the fairness of machine learning models. Key features: - Metrics for assessing which groups are negatively impacted by a model and comparing multiple models in terms of fairness and accuracy metrics. - Algorithms for mitigating unfairness in various AI tasks and fairness definitions. Tech stack: Python Why it matters: Fairlearn provides a comprehensive solution for developers to assess and mitigate unfairness in machine learning models, ensuring fair treatment of all groups in AI systems.

Overview: This repository contains a collection of tutorials on TensorFlow and deep learning, including introductions to TensorFlow, fully connected neural networks, and convolutional neural networks. It serves as an online course taught in 2016. Key features: - TensorFlow tutorials - Introduction to neural networks - Convolutional neural networks Tech stack: Python Why it matters: This repository provides a practical guide for beginners and intermediate learners to understand and implement deep learning concepts using TensorFlow, making it valuable for anyone interested in the field of artificial intelligence and machine learning.

### Overview: nanoMuse is an open-source personal agent designed for every device you own. It's an agent with a name and a unique look, created by Meta, known as [Muse]. nanoMuse does tasks instead of answering questions, stays active even when the app is closed, remembers you, and stops to ask before anything you can't undo. ### Key Features: - **Personal Agent**: A dedicated agent for each device. - **Open Source**: Free, open-source, and non-profit. - **GPL-3.0-or-later License**: Ensures freedom and community collaboration. - **Community Relay**: Runs on a server, allowing you to use the agent without leaving your home. - **Latest Version**: 1.0.5 Lookout, with release notes available. ### Tech Stack: Swift ### Why It Matters: nanoMuse empowers users to automate tasks and stay connected with their devices, making daily life more efficient and personalized. It's designed to be accessible and useful, without any unnecessary complexity.

### Overview: Genann is a minimal, well-tested library for training and using feedforward artificial neural networks in C. It is designed to be simple, fast, reliable, and hackable. ### Key Features: - **C99 with no dependencies**: Genann is written in C99 and does not require any external libraries. - **Single file**: The library is contained in a single source code and header file. - **Simple**: The library is designed to be simple and easy to use. - **Fast and thread-safe**: Genann is fast and thread-safe, making it suitable for real-time applications. - **Easily extendible**: The library is designed to be easily extensible, allowing for easy addition of new features. - **Backpropagation training**: Genann implements backpropagation training, making it suitable for training feedforward neural networks. - **Compatible with alternative training methods**: Genann is compatible with alternative training methods, such as classic optimization and genetic algorithms. - **Examples and test suite**: The library includes examples and a test suite to demonstrate its usage and functionality. ### Tech Stack: C ### Why it Matters: Genann is designed to be simple, fast, reliable, and hackable, making it suitable for a wide range of applications, from basic training to complex neural network architectures. Its self-contained nature and minimal dependencies make it easy to integrate into existing projects.

Overview: This repository provides a comprehensive course on deep reinforcement learning, starting from Markov Decision Processes (MDPs) and tabular reinforcement learning, progressing through advanced topics such as planning, modern AI stacks, world models, and meta-learning. The curriculum is designed to be self-contained and includes a hands-on approach with guided `TODO` sections. Key features: - Foundations: MDPs, tabular RL, DQN, REINFORCE, actor-critic methods, DDPG, TD3, SAC, PPO - Breaking assumptions: RND curiosity, multi-agent RL, offline RL with BC and IQL - Planning: Monte Carlo Tree Search, self-play, AlphaZero-style policy/value learning - Modern AI stack: RLHF with PPO, DPO, GRPO, Decision Transformers, NanoVLA (`DTVLA`) - Production: TensorBoard, checkpointing, debugging, multiple seeds, Ray, Optuna - World models: MBPO with SAC, Dreamer with RSSM latent imagination - Meta + wrap-up: MAML, FOMAML, fast adaptation, and course conclusion Tech stack: Jupyter Notebook Why it matters: This repository offers a practical, hands-on learning experience in deep reinforcement learning, covering foundational concepts and advanced techniques, making it ideal for developers looking to enhance their skills in this field. Repository: alessiodm/drl-zh Primary language: Jupyter Notebook GitHub stars: 2294 GitHub description: Deep Reinforcement Learning: Zero to Hero! README excerpt: # Deep Rein

Overview: This repository contains a comprehensive collection of the best papers, codes, and interpretations from the International Conference on Computer Vision (ICCV) from 1987 to 2019. The repository serves as a resource for developers and researchers interested in the latest advancements in computer vision. Key features: - Papers from all ICCV conferences (ICCV, CVPR, NIPS) from 1987 to 2019. - Codes for selected papers. - Interpretations and discussions of the papers. - Live coding sessions and video recordings from the 2021 ICCV conference. Tech stack: The repository is primarily written in Python and includes Markdown, Git, and various tools for organizing and sharing content. Why it matters: This repository provides a valuable resource for developers and researchers looking to stay updated on the latest advancements in computer vision. It offers a curated collection of papers, codes, and interpretations, making it an essential tool for anyone interested in the field.

Overview: This repository provides a clear, concise, and simple yet powerful and efficient API for deep learning. Key features: - No new issues, pull requests, or commits can be made. - The code remains available for viewing and cloning. - All existing functionality is preserved. Tech stack: Jupyter Notebook Why it matters: This repository offers a straightforward and efficient API for deep learning developers, making it easy to integrate and use in their projects.

### Overview: Agent Teams AI is a comprehensive AI platform designed for teams, allowing you to focus on the big picture while your agents handle the tasks. It includes features like a Kanban board for project management, message and task communication, and integration with various AI models and providers. ### Key Features: - **Kanban Board**: Track project progress and manage tasks efficiently. - **Agent Management**: Assign tasks to agents, monitor their work, and review their outputs. - **AI Model Integration**: Connect with Codex, Claude, OpenCode, Cursor, SuperGrok, GitHub Copilot, Z.AI, MiniMax, Kiro, and more for enhanced productivity. - **Code Review**: Implement code reviews with detailed controls for file-level and hunk-level edits. ### Tech Stack: TypeScript ### Why It Matters: Agent Teams AI simplifies the management of AI agents, allowing teams to focus on their core work while the agents handle the tasks. The platform integrates seamlessly with a variety of AI models and providers, making it a versatile solution for coding and project management. Repository: 777genius/agent-teams-ai Primary Language: TypeScript GitHub Stars: 2287

### Overview: Feature-engine is a Python library designed for feature engineering and selection, compatible with Scikit-learn. It provides a set of transformers to learn and transform features from data, making it an essential tool for machine learning model development. ### Key features: - **Feature Selection**: Filters, selectors, and wrappers for selecting relevant features. - **Feature Engineering**: Transformers for creating new features from existing data. - **Scikit-learn Compatibility**: Works seamlessly with Scikit-learn pipelines and models. ### Tech stack: Python ### Why it matters: Feature-engine simplifies the process of feature engineering and selection, making it easier to build and optimize machine learning models. Its compatibility with Scikit-learn ensures seamless integration with existing workflows and libraries, enhancing the efficiency and effectiveness of feature engineering tasks.

Overview: τ-Bench is a benchmark designed to evaluate tool-agent-user interactions in real-world domains, focusing on collaboration and communication between agents and users. It includes six languages, full-duplex voice, and judges to assess naturalness, speech fidelity, and language-specific checks. Key features: - Six languages: English, Spanish, Portuguese, Hindi, Chinese, and Korean. - Voice in every language: Full-duplex voice runs with real-time agents. - New domain: τ-Entity captures entities on outbound calls. - `banking_knowledge` in voice. - Judges score naturalness, speech fidelity, tool use, and language-specific checks. Tech stack: Python. Why it matters: τ-Bench provides a comprehensive benchmark for evaluating AI agents in real-world collaboration scenarios, ensuring that agents can effectively communicate and collaborate with users in diverse languages and domains. Repository: sierra-research/tau2-bench GitHub stars: 2280 GitHub description: τ-Bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains