Projects

Overview: Caiss是一款高性能的跨平台/多语言的相似向量/相似词/相似句检索引擎，旨在解决各类相似度查询问题，提供全面的增删改查功能，同时支持多种编程语言和环境。 Key features: - 全面的相似度查询功能 - 支持多种编程语言（C++, Python, Java, C#） - 高效的训练和查询性能 - 支持缓存和多线程调用 - 提供C风格的SDK接口 - 支持MacOS, Linux, Windows系统 - 详细的Demo示例和Docker镜像 Tech stack: C++ and Python Why it matters: Caiss旨在解决在人工智能技术普及背景下，海量高维度向量的相似度查询技术在研究和生产中的重要性问题。它通过提供面向最终结果的训练方法和常用距离的训练和查询方式，解决了训练模型结果偏差、距离支持有限、标签信息和向量分离等问题。此外，Caiss还提供了全面的增删改查功能，以及多种语言和环境的支持，使其在各种应用场景中都具有很高的实用价值。 GitHub stars: 549

Overview: This repository contains example notebooks using Haystack, a production-ready open-source AI framework. Key features: - Various model providers - Vector databases - Retrieval techniques - New experimental features Tech stack: Jupyter Notebook Why it matters: These notebooks serve as guidelines for using Haystack with different model providers, vector databases, retrieval techniques, and experimental features, making it a valuable resource for developers looking to integrate Haystack into their projects.

Overview: PySwip is a Python-Prolog interface that allows querying SWI-Prolog in Python programs, featuring an SWI-Prolog foreign language interface, a utility class for Prolog queries, and a Pythonic interface. Key features: - SWI-Prolog foreign language interface - Utility class for easy Prolog querying - Pythonic interface (experimental) Tech stack: Python, SWI-Prolog, ctypes Why it matters: PySwip simplifies integrating Prolog into Python applications, making it easier to leverage Prolog's capabilities within Python projects.

Overview: A person working in the field of AI who doesn’t know math is like a politician who doesn’t know how to persuade. Both have an inescapable area to work upon! Key features: - Algebra - Linear Algebra - Scalars - Vectors - Matrices - Tensors - Eigenvectors & Eigenvalues - Singular Value Decomposition Tech stack: Python Why it matters: This guide is an absolute life-saver for beginners, so you can study the topics that matter most, and an even better resource for practitioners such as myself who require a quick breeze-through on these concepts.

Overview: This project introduces an attentive generative adversarial network designed specifically for removing raindrops from single images, addressing the challenge of visualizing raindrop removal effectively. Key features: - Attentive Generative Adversarial Network (AGAN) - Adversarial Training - Visual Attention Injection - Quantitative and Qualitative Performance Superior to State-of-the-Art Methods Tech stack: Python Why it matters: The AGAN approach effectively addresses the problem of visualizing raindrop removal, demonstrating significant improvements in both quantitative and qualitative metrics compared to existing methods. Repository: rui1996/DeRaindrop Primary language: Python GitHub stars: 548 GitHub description: Attentive Generative Adversarial Network for Raindrop Removal from A Single Image (CVPR 2018)

Overview: LiViM is a real-time video magnification tool that enhances subtle motion and color changes in live webcam or video footage. Key features: - Three magnification algorithms: Motion (Laplace), Motion (Phase), and Color - Frame-accurate timeline, manual playback speed control, and comparison views - Capture via V4L2, Media Foundation, and AVFoundation - Light/dark theme switching based on the operating system - Prebuilt packages available for Linux, Windows, and macOS Tech stack: Qt 6 GUI Why it matters: LiViM provides a powerful way to visualize subtle changes in video footage, making it useful for analyzing motion and color variations in real-time. Repository: tschnz/Live-Video-Magnification GitHub stars: 547

Overview: Happy Transformer simplifies the process of fine-tuning and performing inference with NLP Transformer models. Key features: - Text Generation - Text Classification - Word Prediction - Question Answering - Text-to-Text - Next Sentence Prediction Tech stack: Python Why it matters: Happy Transformer streamlines the fine-tuning and inference process for NLP Transformer models, making it easier for developers to experiment with and deploy state-of-the-art NLP models.

### Overview: CenterNet-better is an improved version of the CenterNet model, designed to be more understandable and performant. It leverages the simplicity of Object as Points (OaP) paper and uses a more elegant implementation, resulting in better performance and reduced training time. ### Key features: - **Simplified Configuration**: Treats configuration files as objects, making it easier to debug. - **Default Training and Testing Scripts**: Includes default scripts for training and testing, requiring only project-specific configs and network modules. - **Automatic Performance Report**: Generates a markdown file automatically after training, providing performance metrics. - **Vectorized Operations**: Improves speed and efficiency through vectorization. ### Tech stack: Python ### Why it matters: This repository aims to simplify and enhance the performance of the CenterNet model, making it more accessible and effective for various applications.

### Overview: This comprehensive deep learning tutorial is designed for beginners and advanced learners alike, offering a step-by-step guide through the world of neural networks and artificial intelligence. ### Key Features: - **Basics of Deep Learning:** Introduces fundamental concepts and algorithms. - **Hands-On Projects:** Includes interactive coding exercises and projects. - **Latest Research:** Covers the latest advancements in deep learning. - **Community Engagement:** Facilitates discussions and collaboration with learners. ### Tech Stack: Jupyter Notebook ### Why It Matters: This tutorial provides a holistic approach to deep learning, combining theoretical knowledge with practical applications, making it an invaluable resource for anyone looking to learn and grow in the field of artificial intelligence. ### GitHub Stars: 547 ### Repository: PeiranLi0930/Comprehensive_DL_Tutor ### Primary Language: Jupyter Notebook ### Welcome to Your Deep Learning Odyssey! 🚀 Embark on a transformative journey into the world of deep learning with "Comprehensive Deep Learning Tutorial: From Zero To Hero". This meticulously crafted tutorial is your personal guide through the fascinating universe of neural networks and AI innovation. Whether you're just starting or aiming to refine your expertise, this repository is your beacon in the expansive deep learning landscape. ## Why This Tutorial? 🤔 - **Holistic Approach:** Integrates coding fundamentals, algorithmic understanding, and insights into the latest research. - **Bite-Sized Learning:** Complex concepts broken down into manageable, understandable segments. - **Hands-On Experience:** From theory to practice, engage with interactive projects

Overview: CLIPort is an end-to-end imitation-learning agent designed to learn a single language-conditioned policy for various tabletop tasks. It combines the broad semantic understanding of CLIP with the spatial precision of TransporterNets to achieve generalizable skills from limited training demonstrations. Key features: - Language-conditioned policy learning - Generalization from limited demonstrations - Integration of CLIP and TransporterNets Tech stack: Python, Jupyter Notebook Why it matters: CLIPort enables robots to learn from minimal human demonstrations, making it particularly useful for tasks requiring precise spatial manipulation. Repository: cliport/cliport GitHub stars: 547 GitHub description: CLIPort: What and Where Pathways for Robotic Manipulation

Overview: This project is a reference architecture for building a travel application using GPT3, providing a seamless user experience with intelligent travel recommendations. Key features: - GPT3-powered travel advice generation - Integration with popular travel APIs - User-friendly interface for booking and managing travel plans Tech stack: TypeScript Why it matters: This project showcases how GPT3 can be utilized to enhance travel applications, offering personalized and intelligent travel recommendations to users.

Overview: 本项目是一款基于u-net、cv2以及cnn的中文车牌定位、矫正和端到端识别软件，采用tensorflow的keras实现，适用于各种拍摄角度和环境条件下的车牌识别任务。 Key features: - 使用u-net进行车牌定位和矫正 - 利用cv2进行车牌区域的边缘检测 - 应用卷积神经网络（cnn）进行车牌的端到端识别 Tech stack: TensorFlow Keras Why it matters: 本项目能够有效识别各种拍摄角度和环境条件下的中文车牌，包括倾斜、强曝光或昏暗环境等，甚至能够识别百度AI车牌识别未能识别的图片，识别效果显著。

### Overview: *faced* is a near real-time CPU-based face detection system utilizing deep learning techniques. It leverages two deep neural networks to achieve robust face detection with high accuracy and real-time performance on CPUs. ### Key Features: - **Real-Time Face Detection**: Detects faces in near real-time on CPUs. - **Ensemble of Neural Networks**: Combines a fully convolutional neural network (FCNN) and a standard CNN for improved accuracy. - **Lightweight Implementation**: Uses TensorFlow, making it suitable for CPUs. - **Robustness Comparison**: Demonstrates superior performance compared to Haar Cascades. ### Tech Stack: - Python - TensorFlow ### Why It Matters: *faced* demonstrates the feasibility of using lightweight neural networks for real-time face detection, showcasing that simpler models can achieve comparable performance to more complex ones, making it a valuable tool for developers and researchers in computer vision and facial recognition.

### Overview: This repository serves as a comprehensive resource for researchers and practitioners interested in Multimodal Retrieval-Augmented Generation (RAG). It collects and categorizes papers related to the field, providing a valuable tool for understanding the latest advancements and their applications. ### Key features: - **Papers**: A curated collection of papers on Multimodal Retrieval-Augmented Generation. - **Taxonomy**: A detailed taxonomy of recent advances and enhancements in the field. - **Application Domains**: A classification of papers based on their application domains. - **Datasets**: Information on popular datasets used in the field. ### Tech stack: - **Programming Languages**: Unknown - **Tools**: GitHub, Git, Markdown, LaTeX ### Why it matters: This repository is essential for researchers and practitioners in the field of Multimodal Retrieval-Augmented Generation, providing a comprehensive overview of the latest advancements and their applications.

### Overview: NÜWA is a state-of-the-art attention network designed for text-to-video synthesis, implemented in PyTorch. This project builds upon the innovative NÜWA model, which has achieved significant progress in the field of text-to-video generation. The extension to video and audio generation using a dual decoder approach makes it particularly versatile. ### Key Features: - **Text-to-Video Synthesis**: NÜWA excels in converting textual descriptions into dynamic videos. - **Dual Decoder Approach**: Extends the model to handle video and audio generation using a dual decoder architecture. - **PyTorch Implementation**: Utilizes PyTorch for efficient and flexible model training and inference. ### Tech Stack: - **Programming Language**: Python - **Model**: NÜWA, a text-to-video synthesis model ### Why It Matters: NÜWA represents a significant advancement in text-to-video synthesis, offering a more efficient and effective approach compared to existing methods. Its ability to generate videos from textual descriptions makes it highly valuable for applications such as content creation, educational tools, and more. ### Repository: lucidrains/nuwa-pytorch - **GitHub Stars**: 547 - **Description**: Implementation of NÜWA, a state-of-the-art attention network for text to video synthesis, in Pytorch. It also includes extensions for video and audio generation using a dual decoder approach. ### Install: ```bash $ pip install nuwa-pytorch ``` ### Usage: 1. **Train the VAE**: ```python

### Overview: XCube is a generative model designed for generating high-resolution sparse 3D voxel grids with arbitrary attributes. It achieves this by employing a hierarchical voxel latent diffusion model that generates grids in a coarse-to-fine manner, significantly reducing test-time optimization time. ### Key Features: - **Hierarchical Voxel Latent Diffusion Model:** Generates grids in a coarse-to-fine manner using a custom framework on VDB data structure. - **Fine-Grained Control:** Supports unconditional and conditional generation, including user-guided editing, scene completion, and text-to-3D. - **High Resolution:** Capable of generating millions of voxels with a finest effective resolution of up to \(1024^3\) without time-consuming optimization. ### Tech Stack: Python ### Why it Matters: XCube offers a significant advancement in 3D generative modeling by enabling high-resolution sparse voxel grids with fine control over attributes, making it highly effective for large-scale applications such as outdoor scene generation and large-scale scene reconstruction.

### Overview: PosterCraft is a unified framework for generating high-quality aesthetic posters. This project aims to revolutionize the aesthetic poster generation process by providing a comprehensive and efficient solution. ### Key Features: - **Unified Framework**: PosterCraft offers a single, unified approach to generate high-quality aesthetic posters. - **Poster100K Dataset**: The project includes a large dataset of 100,000 posters for training and evaluation. - **Models and Datasets**: Users can access models and datasets on HuggingFace, including the Poster100K dataset. - **Live Demo**: A live demo is available on the website, showcasing the project in action. - **GitHub Repository**: The project is hosted on GitHub, with 546 stars and a description of its acceptance by ICLR 2026. - **Community Integration**: The project has been integrated into ComfyUI, making it accessible to a broader audience. ### Tech Stack: - Python ### Why It Matters: PosterCraft simplifies the process of generating high-quality aesthetic posters, making it accessible to a wider audience and contributing to the field of poster design.

### Overview: This repository contains a comprehensive survey of large language models (LLMs) applied to information retrieval (IR). The survey is based on the paper "Large Language Models for Information Retrieval: A Survey" by Yutao Zhu et al., which was published on arXiv. ### Key features: - **Query Rewriter**: Includes methods for prompting, fine-tuning, and knowledge distillation for LLM-based query rewriting. - **Retriever**: Covers techniques using LLMs to generate search data, enhance model architecture, and improve retrieval systems. - **Re-ranker**: Features LLM-based supervised and unsupervised rerankers, discussions on data augmentation, and insights into LLM limitations. ### Tech stack: Python, Markdown, and Git. ### Why it matters: This survey provides a valuable resource for researchers and practitioners looking to understand and utilize LLMs in IR, offering a critical evaluation of current and emerging techniques in the field.

### Overview: This project is a custom implementation of ViTON, designed to help customers virtually try on different clothing items. It addresses the common issue of difficulty in selecting desirable outfits when shopping online, where customers often struggle to try on clothes in the first attempt. ### Key features: - Virtual try-on functionality - Integration with ViTON clothing line - Customizable user interface - Easy-to-use Gradio interface ### Tech stack: Jupyter Notebook ### Why it matters: This project aims to provide a convenient solution for customers who face challenges in trying on clothes online, making the shopping experience more accessible and enjoyable.

Overview: A simple Drowsiness Detection module for humans using computer vision techniques. Key features: - Drowsiness detection algorithm - Eye tracking and aspect ratio analysis - Real-time video stream monitoring Tech stack: Python Why it matters: This module can help prevent accidents by alerting drivers when they are drowsy, ensuring safer driving conditions.