Projects

### Overview: StreamRAG is a powerful tool that enables developers to create, search, and stream videos using ChatGPT, making it easy to gain insights and summaries from specific videos. ### Key features: - 📚 Upload multiple videos to create a library or collection. - 🔍 Search across these videos and get real-time video responses or compilations. - 🛒 Publish your searchable collection on the ChatGPT store. - 📝 Receive summarized text answers (RAG). - 🌟 Gain key insights from specific videos (e.g., "_Top points from episode 31_"). ### Tech stack: StreamRAG is built using Python and Flask, leveraging the capabilities of VideoDB for video uploads and streaming. ### Why it matters: StreamRAG simplifies the process of integrating video search and streaming capabilities into ChatGPT, making it easier for developers to create engaging and informative chatbots.

### Overview: This repository contains the code and data associated with the paper "Denoising Diffusion Models for Plug-and-Play Image Restoration", presented at the CVPR workshop NTIRE 2023. The code is based on the OpenAI Guided Diffusion and DPIR frameworks. It integrates the traditional plug-and-play method into the diffusion sampling framework, aiming to leverage the generative ability of diffusion models for image restoration tasks. ### Key Features: - **Denoising Diffusion Models for Plug-and-Play Image Restoration**: Implements the proposed DiffPIR method. - **Code and Data**: Contains the code and data for the paper. - **Model Download**: Links to model checkpoints can be found in the `model_zoo` directory. - **Inference**: Includes scripts for inference on various image restoration tasks such as super-resolution, image deblurring, and inpainting. - **Training**: Provides instructions on how to train your own diffusion models using the OpenAI Guided Diffusion framework. ### Tech Stack: Python ### Why it Matters: This repository introduces a novel approach to image restoration by integrating diffusion models into the traditional plug-and-play method. The proposed DiffPIR method achieves state-of-the-art performance on FFHQ and ImageNet datasets, demonstrating that diffusion models can serve as effective generative denoiser priors for plug-and-play image restoration tasks.

Overview: This is a PyTorch re-implementation of CPN, a multi-person pose estimation model that won the MSCOCO keypoints2017 challenge. The repository includes evaluation results on the COCO minival dataset and instructions for training and testing various model configurations. Key features: - PyTorch implementation - Evaluation results on COCO minival dataset - Fine-tuned models available for training and testing Tech stack: Python Why it matters: This repository provides a practical tool for evaluating and fine-tuning multi-person pose estimation models, making it valuable for researchers and developers working in the field of computer vision and pose estimation. Repository: GengDavid/pytorch-cpn GitHub stars: 502

Overview: This repository contains the official source code for the paper "F, B, Alpha Matting" by Marco Forte and François Pitié. The project is based on the research presented in the arXiv paper [arXiv:2003.07711](https://arxiv.org/abs/2003.07711). Key features: - The repository includes the official implementation of the F, B, Alpha Matting model. - It provides scripts for foreground, background, and alpha prediction. - The project is under active development for peer-reviewed publication. - The model has been trained on the Adobe Image Matting Dataset. Tech stack: The project is developed using Jupyter Notebook and includes additional packages for Jupyter notebooks. Why it matters: The F, B, Alpha Matting model offers high accuracy in image matting, making it suitable for various applications in image editing and compositing.

### Overview: Explore the advanced Retrieval-Augmented Generation (RAG) system using Langchain, OpenAI GPTs, and META LLAMA3, enhancing Large Language Models (LLMs) with rich, contextual knowledge. ### Key Features: - **Basic RAG**: Understand the journey of a query through RAG, from user input to the final generated response. - **Advanced RAG Techniques**: Explore the intricate components of an advanced RAG system, from query construction to generation. - **Multi Query Retriever**: Get to grips with the Multi Query Retriever structure, enhancing the retrieval process by selecting the best responses from multiple sources. - **Self-Reflection-RAG**: A RAG that has self-reflection/self-grading on retrieved documents and generations. - **Agentic RAG**: A RAG that has an agentic flow on retrieved documents and generations. - **Adaptive Agentic RAG**: A RAG that has an adaptive agentic flow. - **Corrective Agentic RAG**: A RAG that has a corrective agentic flow on retrieved documents and generations. - **LLAMA 3 Agentic RAG Local**: An LLAMA 3 8B Agent RAG that works locally. ### Tech Stack: Jupyter Notebook ### Why It Matters: Enhance your LLMs with the powerful combination of RAG and Langchain for more informed and accurate natural language generation.

Overview: Manifold is an experimental platform designed for long-horizon workflow automation using teams of AI assistants. Key features: - Agent chat - Image generation - Observability (work in progress) - Pulse - Scheduled Tasks - Workflow editor - Specialist registry - Projects - Integrated tools and MCP support - Prompts, datasets, and experiments playground Tech stack: Go Why it matters: Manifold is an experimental frontier AI platform designed for complex, multi-step workflows over extended periods. Repository: intelligencedev/manifold GitHub stars: 502

### Overview Solutions for **CS231n** course assignments offered by Stanford University (Spring 2021-2025). These are the shortest code solutions available. > Check out the solutions for **[CS224n](https://github.com/mantasu/cs224n)**. They contain more comprehensive explanations than others. ### Key features - **Supports**: Python, Jupyter Notebook, PyTorch, TensorFlow, and more. - **Solutions**: Detailed inline explanations for each question. - **Code**: Concise and commented, making it easy to understand and follow along. ### Tech stack - Python - Jupyter Notebook - PyTorch - TensorFlow - Other libraries and tools ### Why it matters These solutions are particularly useful for students and developers looking to quickly solve CS231n assignments without having to go through the full course material. They provide a quick way to understand and implement key concepts in a practical setting. ### Repository The repository is maintained by [mantasu](https://github.com/mantasu). ### GitHub stars The repository has been starred by 502 users, indicating its popularity and usefulness. ### Primary language The repository is written in Jupyter Notebook. ### GitHub description The repository contains solutions for CS231n course assignments, offering a concise and detailed approach to solving problems in the field of computer science and machine learning.

### Overview: EchoTorch is a Python toolkit designed for experimenting with various flavors of Echo State Networks (ESNs) using PyTorch. It provides a simple and efficient way to create and test ESN models for research purposes. ### Key features: - Ready-to-train models - Pre-built datasets for common ESN tasks - Tools for evaluating and comparing ESN models - Neural network components for ESN and Reservoir Computing - Data transformations specific to ESNs - Various utility classes and functions for ESN and Reservoir Computing ### Tech stack: Python ### Why it matters: As it is based on PyTorch, EchoTorch's layers are designed to be easily integrated into deep architectures for future research and development.

Overview: This project, "GLTR: Giant Language Model Test Room," is designed to detect text generated from large language models like GPT-2. It includes a demo instance accessible at [http://gltr.io](http://gltr.io) and a server for `gpt-2-small` that can be started with `python server.py`. Key features: - Detects text generated from large language models - Provides a demo instance for easy testing - Runs on Python 3.6+ - Supports `gpt-2-small` and `BERT` models Tech stack: TypeScript Why it matters: This project is essential for developers and researchers interested in evaluating the quality and authenticity of text generated by large language models, providing a practical tool for testing and validation.

Overview: This project provides a comprehensive Data and Text Mining workflow for analyzing submissions and comments from any given public subreddit using the Pushshift API. Key features: - Downloads submissions and comments from the Pushshift API - Supports fixed amount and target date downloads - Includes ETL process, NLP pipeline, and chart generation - Uses Python libraries: requests, spaCy, NumPy, pandas, matplotlib, seaborn, and wordcloud Tech stack: Python, requests, spaCy, NumPy, pandas, matplotlib, seaborn, wordcloud Why it matters: This project offers a robust solution for analyzing subreddit data, providing insights into submission and comment distribution, word usage, and more, making it valuable for data enthusiasts and researchers alike.

Overview: This repository contains a collection of notes on deep learning, focusing on the foundational concepts and practical applications of neural networks. The notes are based on the "Deep Learning" book by Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Key features: - Detailed explanations of deep learning concepts and techniques - Comprehensive coverage of neural network architectures and algorithms - Practical examples and case studies to illustrate real-world applications Tech stack: Python, Jupyter Notebook, Markdown Why it matters: These notes provide a valuable resource for developers and researchers looking to deepen their understanding of deep learning and its applications in various fields. Repository: Sophia-11/DeepLearningNotes GitHub stars: 501 GitHub description: "《深度学习》手推笔记~持续更新中" README excerpt: 《深度学习》手推笔记~持续更新中 *Last updated: 2019/07/11* #### Update log * 2020/01/02 * - 更新深度学习符号 * 2020/01/20 * - 更新第一章 上传了图片 * 2020/02/03 * - 更新第二章 上传了图片 * 2020/02/16 * - 更新第三章 上传了图片 * 2020/02/25 * - 更新第四章 * 2020/03/04 * - 更新第五章

Overview: VARAG (Vision-Augmented Retrieval and Generation) is a vision-first RAG engine that integrates both visual and textual data to enhance traditional retrieval and generation systems. It supports various retrieval techniques optimized for different use cases, including text, image, and multimodal document retrieval. Key features: - Supports Simple RAG (with OCR) - Incorporates Vision RAG for integrating visual information - Utilizes ColPali RAG for embedding document pages as images - Implements Hybrid ColPali RAG for enhanced retrieval performance Tech stack: Python Why it matters: VARAG leverages advanced visual and textual data integration to improve the accuracy and relevance of text and image retrieval, making it a valuable tool for applications requiring high-quality context-aware responses.

### Overview: SiamMOT is a region-based Siamese Multi-Object Tracking network that detects and associates object instances simultaneously. ### Key features: - **Region-based Siamese Network**: SiamMOT employs a region-based Siamese architecture for multi-object tracking. - **Motion Modeling**: The network includes a motion model for estimating object movement between frames, enhancing tracking accuracy. - **Variants**: Two variants of the Siamese tracker are presented: one that implicitly models motion and another that explicitly models it. ### Tech stack: Python ### Why it matters: SiamMOT significantly improves online multi-object tracking by incorporating motion modeling, outperforming state-of-the-art methods on various MOT datasets and achieving high efficiency with real-time performance.

Overview: This project focuses on Person Image Synthesis via Denoising Diffusion Model, showcasing advancements in image synthesis techniques. Key features: - Customizable dataset for person image synthesis - Comprehensive comparison with state-of-the-art methods in the field - Detailed documentation and reproducible results - Accessible through Google Colab for easy experimentation Tech stack: Jupyter Notebook Why it matters: This project provides a valuable resource for researchers and developers interested in image synthesis and deep learning techniques, offering a practical demonstration of the model's capabilities and a comprehensive comparison with existing methods.

Overview: This project is a sample generative AI tool designed to evaluate Infrastructure as Code (IaC) and architecture diagrams against AWS Well-Architected best practices. Key features: - Prioritization Framework with Eisenhower Matrix - Cross-pillar impact and trade-off analysis - Ready-to-deploy remediation - Cross-pillar impact and trade-off analysis - Architecture diagrams to generate corresponding IaC templates - Interactive chatbot for personalized guidance Tech stack: TypeScript Why it matters: This tool provides a modern web application for assessing infrastructure code and architecture diagrams against AWS Well-Architected best practices, offering insights and recommendations for improvement. Repository: aws-samples/well-architected-iac-analyzer Primary language: TypeScript GitHub stars: 500

Overview: This project is a Python-based spelling corrector that supports multiple languages, including English, Polish, Turkish, Russian, Ukrainian, Czech, Portuguese, Greek, Italian, Vietnamese, French, and Spanish. It is designed to help improve the accuracy of text input and output. Key features: - Supports multiple languages - Can correct full sentences and individual words - Provides multiple correction candidates for words - Offers an option to use faster correction for specific use cases - Allows customization of word sets with `nlp_data` - Offers an alternative method to generate word frequencies using Wikipedia text Tech stack: The project is built using Python, utilizing libraries such as `nlp` for natural language processing and `word_count` for word frequency counting. Why it matters: This spelling corrector is valuable for developers and users who frequently input or output text in languages other than their primary language, ensuring more accurate and consistent text.

### Overview: MaxViT: Multi-Axis Vision Transformer achieves state-of-the-art performance across various vision tasks, including classification, detection, segmentation, image quality, and generative modeling. ### Key features: - **MaxViT Models**: A family of hybrid (CNN + ViT) image classification models that surpass state-of-the-art ConvNets and Transformers in terms of parameter and FLOPs efficiency. - **Global Vision**: MaxViT can "see" globally throughout the entire network, even in earlier, high-resolution stages, due to its linear-complexity grid attention mechanism. - **ImageNet-1k and -21k Results**: Achieves high accuracy on ImageNet-1k and ImageNet-21k datasets. - **Colab Tutorial**: A Google Colab demo is available for running MaxViT on images, providing a practical way to experiment with the model. - **Pretrained Checkpoints**: A list of pre-trained MaxViT checkpoints is provided, including ImageNet-1k, ImageNet-21k, and JFT pre-trained models. ### Tech stack: TensorFlow ### Why it matters: MaxViT models offer improved performance across various vision tasks, making them a valuable addition to the research and development of computer vision systems.

### Overview: Knowledge-driven autonomous driving (AD) has seen significant growth with approaches that emphasize a knowledge-driven paradigm. This repository aims to curate and update resources for this field, providing a comprehensive overview of knowledge-driven AD. ### Key features: - **Dataset & Benchmark**: Various datasets and benchmarks for knowledge-driven AD, including BDD-X, Cityscapes-Ref, DR(eye)VE, HAD, and Talk2Car. - **Environment**: Resources for the environment simulation and testing of knowledge-driven AD systems. - **Driver Agent**: Tools and resources for the driver agent in knowledge-driven AD, such as advice and gaze heatmap. ### Tech stack: None specified. ### Why it matters: Knowledge-driven autonomous driving aims to enhance the decision-making process of self-driving vehicles by incorporating knowledge from various sources, such as expert knowledge, data-driven insights, and human feedback. This approach can lead to more informed and safer autonomous driving systems.

### Overview: 课程涵盖了数字图像处理、深度学习、计算机视觉和机器学习等领域的知识，为开发者提供了全面的课程内容。 ### Key features: - 数字图像处理课件 - 深度学习课件 - 计算机视觉课件 - 机器学习与深度学习课程 - 人工智能导论课程 - 课程使用说明和小问卷 ### Tech stack: HTML ### Why it matters: 这些课程是为开发者设计的，涵盖了数字图像处理、深度学习、计算机视觉和机器学习等领域的知识，帮助开发者提升技能。 ### Repository: fei-aiart/courses ### GitHub stars: 500 ### GitHub description: 课件：数字图像处理，深度学习，计算机视觉，机器学习

### Overview: 智能故障诊断和寿命预测期刊（Journals of Intelligent Fault Diagnosis and Remaining Useful Life）旨在提供高质量的学术资源，帮助研究人员快速找到符合研究兴趣的期刊，从而提高论文发表和学术成果的影响力。 ### Key features: 1. **高质量论文推荐**：提供最新、高质量的学术期刊和会议推荐，帮助读者快速定位研究方向。 2. **分区表更新**：定期更新分区表，包括中科院分区、大区升级和降级情况，以及小类分区。 3. **小类详细分区**：提供更详细的期刊分区信息，帮助读者更好地理解期刊的学术影响力。 4. **CCF/CAAI等级**：提供计算机类和软件类学科认可的期刊目录，帮助读者了解学术期刊的学术水平。 5. **预警期刊**：提供中科院预警期刊列表，帮助读者了解学术期刊的学术风险。 6. **强制OA期刊**：推荐一些提供OA（开放获取）的期刊，方便读者获取最新研究成果。 ### Tech stack: - **GitHub**：用于存储和管理项目代码、文档和资源。 ### Why it matters: - 提供高质量的学术资源，帮助研究人员快速找到符合研究兴趣的期刊，提高论文发表和学术成果的影响力。 - 通过分区表和小类分区信息，帮助读者更好地理解期刊的学术影响力和研究方向。 - 提醒读者关注中科院分区、预警期刊和强制OA期刊，帮助他们选择合适的期刊投稿，提高投稿成功率。 ### Repository: liguge/J