Project directory

Projects

Open-source AI projects tracked by the leaderboard, ordered by GitHub stars.

Python 479 stars

POSTECH-CVLab/SCNeRF

[ICCV21] Self-Calibrating Neural Radiance Fields

Overview: This project introduces Self-Calibrating Neural Radiance Fields (SCNeRF), a novel method for camera self-calibration using Neural Radiance Fields (NeRF). The approach learns camera parameters from raw images without the need for calibration objects, improving accuracy and performance. Key features: - Joint learning of scene geometry and accurate camera parameters - Incorporation of photometric consistency - Utilization of Neural Radiance Fields for accurate rendering - Geometric loss function for complex camera models Tech stack: Python Why it matters: SCNeRF offers a significant improvement in camera calibration accuracy, leading to better neural rendering and performance enhancement in NeRF variants. Repository: POSTECH-CVLab/SCNeRF Primary language: Python GitHub stars: 472 GitHub description: [ICCV21] Self-Calibrating Neural Radiance Fields

Unknown 479 stars

guhhhhaa/4675-scifi

chinese NLP corpus of chinese science fiction,chinese science fiction corpus : About 4675 Chinese science fiction novels 大约有4675本科幻小说，中文科幻小说自然语言处理语料库，中文科幻小说文本语料库，中文科幻小说文本数据库，科幻小说语料

Overview: This repository contains a Chinese NLP corpus of 4,675 Chinese science fiction novels, which serves as a natural language processing corpus for Chinese science fiction novels. Key features: - A collection of 4,675 Chinese science fiction novels - Available in both Chinese and English - Suitable for various NLP tasks such as text classification, sentiment analysis, and entity recognition Tech stack: Python, NLTK, spaCy, and other NLP libraries Why it matters: This corpus provides a rich dataset for researchers and developers working on natural language processing tasks related to Chinese science fiction, enabling better understanding and improvement of NLP models in this specific domain. Repository: guhhhhaa/4675-scifi Primary language: Chinese GitHub stars: 479

Python 479 stars

paulpierre/markdown-crawler

A multithreaded 🕸️ web crawler that recursively crawls a website and creates a 🔽 markdown file for each page, designed for LLM RAG

Overview: This is a multithreaded web crawler designed for large language model document parsing, specifically for LLM RAG use cases. It creates markdown files for each page of a website. Key features: - Threading support for faster crawling - Continue scraping where you left off - Set the max depth of children to crawl - Support for tables, images, etc. - Validates URLs, HTML, and filepaths - Configurable markdown heading style - Browser-like User-Agent to bypass JS checks - Uses BeautifulSoup to parse HTML - Verbose logging option - Ready-to-go CLI interface - Comprehensive test suite (95% coverage) Why it matters: This tool is essential for simplifying the process of parsing and organizing large documents for LLM RAG, making it easier to extract and process information efficiently. Repository: paulpierre/markdown-crawler Primary language: Python GitHub stars: 479 GitHub description: A multithreaded 🕸️ web crawler that recursively crawls a website and creates a 🔽 markdown file for each page, designed for LLM RAG

Go 479 stars

takara-ai/go-attention

A full attention mechanism and transformer in pure go.

### Overview: This project introduces a pure Go implementation of attention mechanisms and transformer layers, designed for high performance, consistency, and production reliability. It addresses common concerns about Go in AI/ML, such as SIMD support, memory efficiency, parallel performance, and numerical stability. ### Key Features: - **Attention Mechanisms**: Implementations of dot product, scaled dot product, and multi-head attention. - **Transformer Layers**: Support for encoder and decoder layers, including positional encoding and layer normalization. - **Efficient Performance**: Single optimized code path ensures consistent, predictable performance across all input sizes. - **Edge-Optimized**: Zero external dependencies and minimal memory footprint perfect for edge devices. - **Production-Ready**: Comprehensive error handling, type safety, and deterministic behavior. - **Scalability**: Efficient batched operations support high-throughput cloud deployments. ### Tech Stack: Go ### Why It Matters: - **Performance Without Compromise**: Delivers production-grade performance for AI workloads. - **Addressing Go's AI Limitations**: Solves common concerns about Go in AI/ML. - **Real-World Benefits**: Eliminates dependency resolution delays, consistent performance across all hardware, and cost-effective resource usage. - **Core Types**: Provides clear definitions for `Vector` and `Matrix` types for easy integration. ### API Documentation: - **Core Types**: `Vector` and `Matrix` types for representing vectors and matrices. - **Quick Examples**: Demonstrates basic and multi-head attention mechanisms, showcasing the simplicity and effectiveness of the implementation. ### Performance Characteristics:

C++ 478 stars

D7EAD/liboai

A C++17 library to access the entire OpenAI API.

Overview: liboai is a C++17 library designed to access the entire OpenAI API, offering a simplified interface for developers. Key features: - ChatGPT - Audio - Azure - Functions - Image DALL·E - Models - Completions - Embeddings - Files - Fine-tunes - Moderation - Asynchronous Support Tech stack: C++ Why it matters: liboai provides a straightforward and intuitive way for developers to interact with OpenAI's API, making it accessible and easy to integrate into various projects.

Python 478 stars

Haochen-Wang409/U2PL

[CVPR'22 & IJCV'24] Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels & Using Unreliable Pseudo-Labels for Label-Efficient Semantic Segmentation

### Overview: This project introduces an effective pipeline for semi-supervised semantic segmentation using unreliable pseudo-labels. The approach leverages the entropy of predictions to separate reliable and unreliable pixels, treating them as negative samples for training. This method significantly improves the model's performance on various benchmarks, including PASCAL VOC 2012. ### Key Features: - **Entropy-Based Pixel Classification:** Separates reliable and unreliable pixels using the entropy of predictions. - **Negative Sampling:** Treats unreliable pixels as negative samples to improve model training. - **Adaptive Thresholding:** Adjusts the threshold for reliable-unreliable partition based on the training evolution. - **Experimental Validation:** Demonstrates superior performance over state-of-the-art alternatives on various benchmarks. ### Tech Stack: - Python ### Why It Matters: This method addresses a common challenge in semi-supervised semantic segmentation by effectively utilizing unreliable pseudo-labels, thereby improving model robustness and performance.

Python 478 stars

LituRout/RF-Inversion

Rectified Flow Inversion (RF-Inversion) - ICLR 2025

Overview: Rectified Flow Inversion (RF-Inversion) is a novel approach for image inversion and editing using rectified stochastic differential equations. This method efficiently inverts reference style images without requiring text descriptions and applies desired edits based on new prompts, ensuring semantic image editing and stylization. Key features: - Efficient image inversion and editing without text descriptions. - Supports semantic image editing and stylization based on prompts. - Inverts reference content images and applies desired edits. - Supports Gradio and Hugging Face demos for better user experience. Tech stack: Python Why it matters: This method offers a new and efficient way to manipulate and edit images, making it useful for various applications such as content modification, style transfer, and creative design. Repository: LituRout/RF-Inversion Primary language: Python GitHub stars: 478

TypeScript 478 stars

NPC-Worldwide/incognide

Explore the unknown, build the future, own your data.

Overview: Incognide is a versatile workspace that integrates chat, code, documents, web browsing, and media into a tileable workspace, eliminating distractions and context switching. Key features: - Seamless integration of chat, code, documents, web browsing, and media - Intelligent context and composable automations - Path-based organization to keep work structured naturally - Auto-saving workspaces for easy resume of projects - Supports Linux, macOS, and Windows platforms Tech stack: TypeScript Why it matters: Incognide simplifies workflows by eliminating distractions and context switching, making it ideal for developers and professionals who need to work efficiently and focus on their tasks. Repository: NPC-Worldwide/incognide GitHub stars: 477 GitHub description: Explore the unknown, build the future, own your data.

Jupyter Notebook 478 stars

abdur75648/Deep-Learning-Specialization-Coursera

This repo contains the updated version of all the assignments/labs (done by me) of Deep Learning Specialization on Coursera by Andrew Ng. It includes building various deep learning models from scratch and implementing them for object detection, facial recognition, autonomous driving, neural machine translation, trigger word detection, etc.

### Overview: This repository contains all the assignments and labs from Coursera's Deep Learning Specialization by Andrew Ng, updated for the latest developments in deep learning and programming frameworks. It includes building various deep learning models from scratch, such as object detection, facial recognition, autonomous driving, neural machine translation, and trigger word detection. ### Key Features: - Updated versions of all assignments and labs from Coursera's Deep Learning Specialization. - Includes building various deep learning models from scratch. - Covers object detection, facial recognition, autonomous driving, neural machine translation, and trigger word detection. - Updated for the latest developments in deep learning and programming frameworks. ### Tech Stack: Jupyter Notebook ### Why It Matters: This repository provides a comprehensive resource for learning and practicing deep learning concepts, including building and implementing various models and frameworks. It is useful for developers and learners looking to enhance their skills in deep learning and programming.

Python 478 stars

microsoft/seismic-deeplearning

Deep Learning for Seismic Imaging and Interpretation

### Overview: This repository provides a comprehensive guide for performing seismic imaging and interpretation on Azure, utilizing state-of-the-art deep learning techniques. ### Key Features: - Sample notebooks demonstrating how to use the codebase. - Experiments that run Python scripts to train and test machine learning models. - Utilities for reproducible and scalable machine learning tasks. - Sample notebooks and experiments organized for easy access. - Docker-enabled for quick and easy demo setup. ### Tech Stack: Python ### Why It Matters: This repository empowers geophysicists and data scientists to perform advanced seismic imaging and interpretation tasks on Azure, leveraging cutting-edge deep learning models and techniques. ### Repository: microsoft/seismic-deeplearning ### Primary Language: Python ### GitHub Stars: 478 ### GitHub Description: This repository shows you how to perform seismic imaging and interpretation on Azure. It empowers geophysicists and data scientists to run seismic experiments using state-of-art DSL-based PDE solvers and segmentation algorithms on Azure. ### Quick Start: - **Docker**: Quick start guide for Docker-enabled environment. - **Hands-On**: - Review the demo notebook for an idea of how the models are used. - Set up a compute environment and download datasets for running the code. ### Azure Machine Learning: Azure Machine Learning enables training and deployment of machine learning models and pipelines at scale, leveraging open-source Python frameworks. ### Interpretation: For seismic interpretation, the repository consists of extensible machine learning pipelines, showcasing how to leverage state-of

Python 477 stars

Evil0ctal/Fast-Powerful-Whisper-AI-Services-API

⚡ 一款用于自动语音识别 (ASR)、翻译的高性能异步 API。不需要购买Whisper API，使用本地运行的Whisper模型进行推理，并支持多GPU并发，针对分布式部署进行设计。还内置了包括TikTok、抖音等社交媒体平台的爬虫，可实现来自多个社交平台的无缝媒体处理，为媒体内容数据自动化处理提供了强大且可扩展的解决方案。

### Overview: This project aims to provide a powerful and ready-to-use Whisper API service, designed for high performance, scalability, and distributed processing requirements. Built around a producer-consumer model, it is ideal for large-scale, highly efficient automatic speech recognition scenarios. The project leverages the OpenAI Whisper model and the Faster Whisper model, delivering faster inference with nearly equivalent accuracy. It supports high-quality multilingual transcription and translation tasks, and includes an integrated crawler module for seamless media processing from social media platforms like Douyin and TikTok. ### Key Features: - **Asynchronous Design**: Built on Python 3.11’s asyncio capabilities, all modules are asynchronous, enabling efficient request handling and enhancing overall system stability and high concurrency. - **Built-in Document UI**: Leverages FastAPI’s auto-generated OpenAPI JSON, including an interactive Swagger UI for visual API testing in the browser. Swagger UI provides bilingual descriptions in English and Chinese with default parameter settings, enabling quick start testing for users. - **High Accuracy**: The latest `large-v3` model ensures accurate output, and Faster Whisper significantly reduces inference time while maintaining high accuracy. - **Distributed Deployment**: The project can access and store tasks in a shared database and plans seamless integration with Kafka, achieving a harmonious integration of FastAPI and Kafka for building real-time, intelligent Web APIs. - **Asynchronous Model Pool**: Implements an efficient asynchronous AI model pool that supports multi-instance concurrent processing for OpenAI Whisper and Faster Whisper models under thread-safe conditions. In CUDA-accelerated, multi-GPU environments, intelligent loading

Jupyter Notebook 477 stars

Guitaricet/relora

Official code for ReLoRA from the paper Stack More Layers Differently: High-Rank Training Through Low-Rank Updates

### Overview: ReLoRA is an efficient method for training large language models by leveraging existing LoRA parameters. It allows for the integration of LoRA into the main model, enabling faster training and better performance. ### Key Features: - **Relora Integration**: Combines LoRA parameters with the main model, reducing the need for retraining. - **Reset Frequency**: The reset frequency is determined by the `--relora` parameter, which specifies the number of update steps, not global steps. - **Training Script**: Provides a 1B training script that includes warm-up and ReLoRA stages, optimizing learning rates and batch sizes. ### Tech Stack: - **Python 3.10+** - **PyTorch 2.0+** - **Flash Attention** ### Why It Matters: ReLoRA is particularly beneficial for models like LLaMA, where the warm-up phase can be time-consuming. By leveraging existing LoRA parameters, ReLoRA can significantly reduce the training time, making it more efficient and faster to train large language models.

Python 477 stars

Lakonik/LakonLab

Official implementation of AsymFlow, pi-Flow, GMFlow

Overview: LakonLab is an official PyTorch implementation of AsymFlow, pi-Flow, and GMFlow, providing comprehensive tools for flow-based image generation and manipulation. Key features: - AsymFlow: A flow-based model for image generation and manipulation. - pi-Flow: A policy-based flow model for image generation and manipulation. - GMFlow: A Gaussian Mixture Flow matching model for image generation and manipulation. Tech stack: Python Why it matters: LakonLab offers a robust and versatile toolkit for researchers and developers interested in flow-based image processing and generation, providing a solid foundation for various applications in computer vision and AI. Repository: Lakonik/LakonLab GitHub stars: 477 GitHub description: Official implementation of AsymFlow, pi-Flow, and GMFlow README excerpt: Official PyTorch implementation of the papers: AsymFlow, pi-Flow, and GMFlow. Includes detailed documentation and examples for using the models.

Python 477 stars

inseq-team/inseq

Interpretability for sequence generation models 🐛 🔍

Overview: This project focuses on interpretability for sequence generation models, providing valuable insights into the generated sequences. Key features: - Model interpretability tools - Visualization of model predictions - Code documentation and examples Tech stack: Python Why it matters: Understanding why a model generated a specific sequence can be crucial for applications requiring transparency and accountability, such as in finance or healthcare.

Go 477 stars

james-bowman/nlp

Selected Machine Learning algorithms for natural language processing and semantic analysis in Golang

### Overview: James Bowman's `nlp` repository is a comprehensive collection of selected machine learning algorithms for natural language processing and semantic analysis in Go. This package leverages the Gonum library for linear algebra and scientific computing, inspired by Python's scikit-learn and Gensim. It includes implementations of Latent Semantic Analysis (LSA), Fast SimHash, Random Indexing (RI), Reflective Random Indexing (RRI), Latent Dirichlet Allocation (LDA), Principal Component Analysis (PCA), and TF-IDF weighting. The repository is designed to handle large web-scale corpora efficiently, providing scalable solutions for semantic analysis and retrieval of semantically similar documents. ### Key Features: - **LSA (Latent Semantic Analysis)**: Truncated Singular Value Decomposition for dimensionality reduction. - **Fast SimHash**: Approximate cosine similarity/angular distance comparisons using multi-index and Forest schemes. - **RI (Random Indexing)**: Scalable LSA over large corpora. - **LDA (Latent Dirichlet Allocation)**: Parallelized implementation of SCVB0 algorithm for topic extraction. - **PCA (Principal Component Analysis)**: Efficient memory usage and processing over large document corpora. - **TF-IDF**: Weighting to account for frequently occurring words. - **Sparse Matrix**: Efficient memory usage and processing over large document corpora. - **Stop Word Removal**: Removing common English words to improve accuracy. - **Feature Hashing**: Reduced memory requirements and reduced reliance on training data. ### Tech Stack: The repository is built using Go, leveraging

Python 477 stars

njustkmg/OMML

Multi-Modal learning toolkit based on PaddlePaddle and PyTorch, supporting multiple applications such as multi-modal classification, cross-modal retrieval and image caption.

Overview: OMML is a multi-modal learning toolkit built on PaddlePaddle and PyTorch, designed to support various applications such as multi-modal classification, cross-modal retrieval, and image captioning. It aims to provide efficient solutions for processing multi-modal data, enhancing the applications of multi-modal machine learning. Key features: - Multiple task scenarios: OMML offers a variety of multi-modal learning task algorithm model libraries, including multi-modal fusion, cross-modal retrieval, image captioning, and supports user-defined data and training. - Successful industrial applications: The toolkit has been applied in real-world scenarios, such as sneaker authenticity identification and image description. - Visualization: OMML includes a visualization module that allows users to visualize the results of fusion tasks, such as the identification of sneaker authenticity. - Enterprise application: OMML has been used in cooperation with Baidu TIC for smart recruitment resume analysis, demonstrating its effectiveness. - Framework: OMML consists of three modules: data processing, model library, and trainer, providing a unified interface for efficient data processing and model training. Why it matters: OMML simplifies the development process for multi-modal learning tasks, making it easier for developers to integrate and utilize multi-modal data for various applications, thereby accelerating the development of multi-modal machine learning solutions. Tech stack: OMML is built on PaddlePaddle and PyTorch, utilizing the strengths of both libraries to support efficient multi-modal learning tasks. GitHub stars: 477 Repository: njustkmg/OMML

JavaScript 476 stars

BaYue-SYJ/shuixian-prompts

水仙的AI提示词｜5,000+ 中文 AI 绘画提示词库：检索 / 分类 / 一键复制。提示词覆盖 ChatGPT、Grok、Gemini、Midjourney、Stable Diffusion、Flux、即梦、可灵、豆包、海螺、文心一格等主流生图模型。源码 + 数据集全部开源。

Overview: 水仙的AI提示词库是一个包含5,000+条中文AI绘画提示词的集合，覆盖了多种主流生图模型，如Midjourney、Stable Diffusion、Flux、ChatGPT、Gemini、即梦、可灵、豆包、海螺、文心一格等。用户可以通过关键词搜索、标签分面、筛选器等功能轻松找到和使用提示词。 Key features: - 关键词搜索 - 拼音首字母搜索 - 标签分面筛选 - 一键复制提示词 - 图片下载与灯箱浏览 - 备注与“用过”标记 - 收藏合集管理 - 导出Markdown、JSON、CSV - PWA离线浏览与轻量数据 Tech stack: JavaScript Why it matters: 该库为设计师、艺术家和创意工作者提供了一个便捷的工具，帮助他们快速生成高质量的AI绘画作品，无需复杂的代码学习。通过丰富的提示词库，用户可以轻松找到灵感，提高创作效率。

Python 476 stars

PhantomInsights/mexican-government-report

Text Mining on the 2019 Mexican Government Report, covering from extracting text from a PDF file to plotting the results.

Overview: This project involves extracting text from a PDF, cleaning it, and then passing it through an NLP pipeline to analyze the government report from the 2019 Mexican Government Report. Key features: - Extracting text from a PDF file - Cleaning the extracted text - Passing the cleaned text through an NLP pipeline - Presenting the results with graphs Tech stack: Python Why it matters: This project is important because it demonstrates the process of text mining and NLP on a real-world dataset, providing insights into the government report's content and trends.

Zig 476 stars

antflydb/antfly

### Overview: Antfly is an open-source search-and-inference database written in Zig, designed for efficient full-text, dense, and sparse vector indexing, as well as graph traversal. It includes models for chunking, embedding, reranking, OCR, and entity extraction, all running within a single process. ### Key features: - **Hybrid Search**: Full-text (BM25), dense vectors, and sparse vectors, combined with reciprocal rank or relative score fusion. - **Full-text Search**: Lucene-style segments with highlighting, geo, regex, wildcard, and fuzzy queries. - **RAG Agents**: Built-in retrieval-augmented generation agents to improve search results. - **Graph Traversal**: Supports graph traversal over the same table. - **Models**: Models for chunking, embedding, reranking, OCR, and entity extraction. ### Tech stack: Zig ### Why it matters: Antfly is designed for efficient and scalable search and inference, making it ideal for applications requiring fast and accurate data retrieval and analysis. Its modular architecture and in-process execution ensure low latency and high performance, making it suitable for both local development and production environments.

C++ 476 stars

arjo129/uSpeech

Speech recognition toolkit for the arduino

Overview: uSpeech is a speech recognition toolkit designed for Arduino, providing phoneme-based recognition with a focus on accuracy and simplicity. Key features: - Letter-based recognition - Small memory footprint - Arduino-compatible - No training required - Fixed point arithmetic - 30% to 40% accuracy for phoneme-based recognition, up to 80% for words Tech stack: C++ Why it matters: uSpeech offers a straightforward and efficient solution for speech recognition on Arduino, suitable for projects requiring real-time speech processing without the need for complex training. Repository: arjo129/uSpeech GitHub stars: 476 GitHub description: Speech recognition toolkit for the arduino