Projects

### Overview This repository contains the source code for all the projects of the Udacity Self-Driving Car Engineer Nanodegree. It serves as a comprehensive resource for anyone interested in learning about the development of self-driving cars. ### Key Features - **Basic Lane Finding (P1):** Implementing a simple algorithm to detect lane lines in a video stream. - **Traffic Signs (P2):** Developing a model to classify traffic signs from a camera input. - **Behavioral Cloning (P3):** Training a neural network to drive a car based on a video of a skilled driver. - **Advanced Lane Finding (P4):** Enhancing the lane detection algorithm to improve accuracy. - **Vehicle Detection (P5):** Detecting and tracking vehicles in real-time video streams. - **Extended Kalman Filter (P6):** Improving the accuracy of vehicle detection using an extended Kalman filter. - **Unscented Kalman Filter (P7):** Implementing an unscented Kalman filter for more accurate state estimation. - **Kidnapped Vehicle (P8):** Developing a system to track and recover a vehicle that has been kidnapped. - **PID Controller (P9):** Implementing a proportional-integral-derivative controller for vehicle control. - **MPC Controller (P10):** Using model predictive control for more precise vehicle control. ### Tech Stack - **Programming Languages:** C++ - **Frameworks/Tools:** TensorFlow, OpenCV, Python ### Why It Matters This repository provides a practical way

### Overview: Baichuan-13B is a 13B-parameter large language model developed by Baichuan Intelligent Technology, offering both pre-trained and aligned models for developers and general users alike. ### Key Features: - **Pre-trained Model (Baichuan-13B-Base)**: A 130 billion parameter model that excels in both Chinese and English benchmarks. - **Aligned Model (Baichuan-13B-Chat)**: A chat-oriented model with strong conversational capabilities, designed for easy deployment with just a few lines of code. ### Tech Stack: Python ### Why It Matters: Baichuan-13B is a significant advancement in the field of large language models, offering unparalleled performance across various benchmarks and a user-friendly approach to deployment. Its ability to be both pre-trained and aligned makes it highly versatile and accessible to developers and the general public alike.

### Overview: Anomaly detection is a crucial technique used to identify unusual patterns in data that deviate from the expected behavior. This list aims to provide a curated collection of resources and papers related to anomaly detection, including deep learning approaches, time-series, video, and image-based methods. The list is updated regularly and covers various types of anomaly detection, such as novelty detection, out-of-distribution detection, and segmentation. ### Key features: - **Deep Learning for Anomaly Detection**: A survey of deep learning methods for anomaly detection. - **Anomaly Classification Target**: Identifying anomalies in time-series data. - **Out-Of-Distribution Detection Target**: Detecting anomalies in image and video data. - **Anomaly Segmentation Target**: Segmenting abnormal regions in image and video data. ### Tech stack: - Python, TensorFlow, PyTorch, and other machine learning frameworks. ### Why it matters: Anomaly detection is essential in various fields such as cybersecurity, healthcare, finance, and manufacturing, where it helps in identifying fraudulent activities, detecting diseases, detecting financial anomalies, and identifying defective products, respectively.

### Overview: Image and Video Deblurring is a curated list of resources focused on the process of removing blur from images and videos. This includes techniques for single-image motion deblurring, non-blind deblurring, multi-image/Video motion deblurring, depth-aware motion deblurring, defocus deblurring, and benchmark datasets for motion deblurring. ### Key Features: - **Single-Image-Blind-Motion-Deblurring (non-DL):** Techniques for removing camera shake from a single photograph. - **Single-Image-Blind-Motion-Deblurring (DL):** Deblurring using deep learning methods. - **Non-Blind-Deblurring:** Techniques that do not require motion information. - **Multi-image/Video-Motion-Deblurring:** Deblurring of multiple images or videos. - **Depth-Aware Motion Deblurring:** Deblurring that considers depth information. - **Defocus Deblurring and Potential Datasets:** Techniques for removing defocus blur and related datasets. - **Benchmark Datasets on Motion Deblurring:** Resources for evaluating motion deblurring algorithms. - **AI Photo Enhancer Apps:** Applications that enhance images using deblurring techniques. ### Tech Stack: - **Python** - **OpenCV** - **Matplotlib** - **TensorFlow** - **PyTorch** - **Keras** - **Pillow** ### Why It Matters: Deblurring techniques are crucial in enhancing the quality of images and videos, making them suitable for various applications such

Overview: PocketFlow is an open-source framework designed to automatically compress and accelerate deep learning models, making them more efficient for deployment on mobile devices with limited computational resources. Key features: - Automatically compresses and accelerates deep learning models with minimal human effort. - Supports various model compression algorithms for efficient model size and speed. - Provides a toolkit for developers to improve inference efficiency without performance degradation. - Compatible with Python, making it accessible to a wide range of developers. Tech stack: Python Why it matters: PocketFlow simplifies the process of compressing and accelerating deep learning models, enabling faster and more efficient applications on mobile devices, thus expanding the potential of deep learning in various fields.

### Overview: VieNeu-TTS is the next generation of Vietnamese TTS, featuring 10,000+ hours of bilingual training, instant voice cloning, dedicated podcast/conversation mode, and real-time streaming with high audio quality and low latency. ### Key Features: - **10,000+ hours of bilingual training** - **Instant voice cloning** - **Podcast/Conversation mode** - **48kHz audio quality** - **Real-time streaming with low latency (~115 ms)** ### Tech Stack: Python ### Why It Matters: VieNeu-TTS offers advanced text-to-speech capabilities with high fidelity and real-time streaming, making it ideal for applications requiring seamless voice cloning and on-device processing.

### Overview: UltraChat is a comprehensive dataset and model repository focused on large-scale, informative, and diverse multi-round chat data. It includes UltraLM-13B, UltraLM-65B, and UltraRM-13B, among others, with a particular emphasis on the 13B version which ranks highly on the AlpacaEval Leaderboard. ### Key Features: - **UltraLM-13B**: A powerful chat language model trained on UltraChat, ranking #1 among open-source models and #4 among all models on the AlpacaEval Leaderboard. - **UltraLM-65B**: A larger model version, providing even more comprehensive and diverse data. - **UltraRM-13B**: A reward model that significantly improves the win rates of text-davinci-003 on the AlpacaEval benchmark. - **UltraCM-13B**: A critic model that complements UltraRM, further enhancing the model's performance. ### Tech Stack: - Python ### Why It Matters: UltraChat provides a rich and diverse dataset for training and evaluating chat language models, making it a valuable resource for researchers and developers in the field of natural language processing. ### Repository: Repository: thunlp/UltraChat Primary language: Python GitHub stars: 2908 ### GitHub Description: Large-scale, Informative, and Diverse Multi-round Chat Data (and Models) ### Data Release: - **UltraLM-13B**: Releasing a large-scale, fine-grained, diverse preference

### Overview: 使用机器学习算法自动识别12306验证码 ### Key features: - 文字识别模型 - 图片识别模型 - 自动识别12306验证码 ### Tech stack: Python ### Why it matters: 通过机器学习技术，显著提高了12306验证码识别的准确性和效率，为用户提供更加便捷的乘车体验。

### Overview: Deep QA is a Python implementation of a neural conversational model, based on TensorFlow, that aims to reproduce the results of the Google chatbot. It uses a sequence-to-sequence model for sentence predictions and supports various dialog corpora, including the Cornell Movie Dialogs, OpenSubtitles, Supreme Court Conversation Data, Ubuntu Dialogue Corpus, and custom data formats. ### Key features: - **Sequence-to-Sequence Model**: Uses a Recurrent Neural Network (RNN) to predict responses in conversations. - **Support for Multiple Corpora**: Compatible with the Cornell Movie Dialogs, OpenSubtitles, Supreme Court Conversation Data, Ubuntu Dialogue Corpus, and custom datasets. - **Pretrained Word Embeddings**: Utilizes pre-trained word embeddings for improved performance. - **Training and Testing**: Train the model using `main.py`, and test it with `main.py --test` or `main.py --test interactive`. - **Web Interface**: Provides a simple web interface for better interaction and visualization of results. ### Tech Stack: - **Python 3.5** - **TensorFlow** - **Numpy** - **CUDA (for GPU support)** - **NLTK (for tokenizing sentences)** - **Tqdm (for progress bars)** - **Django (for web interface)** - **Channels (for real-time communication)** - **Redis (for session management)** - **asgi_redis (for better Redis integration)** - **Pre-trained Word Embeddings (for faster training)** ### Why it Matters: Deep QA aims to

### Overview: Moltis is a secure, persistent personal agent server written entirely in Rust. It runs on your hardware, providing a single binary that is sandboxed and secure. The agent supports multiple LLMs, voice, memory, and integrates with popular chat apps and tools like Telegram, WhatsApp, Discord, and Teams. Moltis is designed for auditable and secure operation, with a small binary size and a modular crate-based architecture. ### Key features: - **Secure by Design**: Your keys never leave your machine, ensuring complete control and security. - **Your Hardware**: Runs on a Mac Mini, Raspberry Pi, or any server you own, with a single Rust binary. - **Full-featured**: Voice, memory, cross-session recall, automatic edit checkpoints, scheduling, and integration with popular chat apps and tools. - **Auditable**: The agent runner and model interface are small, with 59 modular crates and 470+ Rust files containing tests. - **Comparison**: Moltis uses Rust, whereas OpenClaw and Hermes Agent use TypeScript and Python, respectively. Moltis is a persistent personal agent server with a single Rust binary, while OpenClaw and Hermes Agent are gateway agents with learning loops and research tooling. ### Tech stack: Rust. ### Why it matters: Moltis is designed for auditable and secure operation, with a small binary size and a modular crate-based architecture. It integrates with popular chat apps and tools, making it a powerful and versatile tool for personal agents.

Overview: Texthero is a Python toolkit designed for quick and effortless text-based dataset manipulation, providing a modern and user-friendly interface for programmers of the 2020 decade. Key features: - Text preprocessing - Natural Language Processing (NLP) including keyphrases and keywords extraction, and named entity recognition - Text representation with TF-IDF, term frequency, and custom word embeddings - Vector space analysis with clustering (K-means, Meanshift, DBSCAN, Hierarchical), topic modeling (in development), and interpretation - Text visualization with vector space visualization and place localization on maps (in development) Tech stack: Python Why it matters: Texthero simplifies the process of working with text-based datasets, making it accessible to programmers without extensive linguistic knowledge. It offers a modern and user-friendly interface, making it easy to understand and use, even for those new to NLP. Repository: jbesomi/texthero Primary language: Python GitHub stars: 2905 GitHub description: Text preprocessing, representation and visualization from zero to hero.

### Overview: Implementing popular machine learning and deep learning papers in 100 lines of code. ### Key features: - Maxout Networks - Network In Network - Playing Atari with Deep Reinforcement Learning - Auto-Encoding Variational Bayes - Generative Adversarial Networks - Conditional Generative Adversarial Nets - Adam: A Method for Stochastic Optimization - NICE: Non-linear Independent Components Estimation - Human-level control through deep reinforcement learning - Deep Unsupervised Learning using Nonequilibrium Thermodynamics - Variational Inference with Normalizing Flows - Deep Reinforcement Learning with Double Q-learning - Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks - Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs) - Deep Residual Learning for Image Recognition ### Tech stack: Python ### Why it matters: These papers have significantly impacted the field of machine learning and deep learning, demonstrating the power of concise and efficient implementations.

### Overview: PenguinHarness is a unified and stable RSI (Recursive Self-Improvement) platform designed for agent self-evolution. It offers a single framework that supports various RSI algorithms, including default, OPRO, APE, ACE, and AWM, all running on the same Benchmarks, scoreboard, and snapshots. The framework is lightweight and affordable, ideal for open models. ### Key Features: - **Unified Framework:** Supports multiple RSI algorithms. - **Automatic Optimization:** Any model, any algorithm, any benchmark. - **One Sentence Builds an Agent App:** Describe what you need, and an agent writes the app, its page, and run instructions. - **Cheap to Run:** Deliberately small toolset, optimized for open models, with the best accuracy at 1/70 of Claude Code's cost. ### Tech Stack: TypeScript ### Why It Matters: PenguinHarness simplifies the process of creating agent-based applications, making it accessible and affordable for developers and researchers. Its unified framework and automatic optimization capabilities make it a valuable tool for anyone looking to automate the creation of intelligent agents.

### Overview: This repository contains a collection of survey papers that summarize recent advancements in various fields of machine learning, including deep learning, natural language processing, computer vision, and more. The papers provide valuable insights into the current state of the art and highlight the most impactful research in each area. ### Key Features: - **Deep Learning**: Surveys on architectures, theory, and applications. - **Natural Language Processing**: Reviews of deep learning techniques, classification, generation, and evaluation methods. - **Computer Vision**: Focuses on object detection, adversarial attacks, autonomous vehicles, and image captioning. - **Reinforcement Learning**: Reviews of algorithms and applications in reinforcement learning. - **Graphs**: Surveys on embeddings, meta-learning, and other graph-based techniques. - **Others**: Includes surveys on neural architecture search, neural architecture search, and more. ### Tech Stack: The repository uses Markdown for documentation and includes links to external papers and resources. ### Why It Matters: These survey papers are essential for developers and researchers looking to stay updated with the latest trends and methodologies in machine learning. They provide a comprehensive overview of the field, helping readers understand the current landscape and identify areas where they can contribute or innovate.

Overview: Euler-2.0 is a distributed graph deep learning framework designed for efficient processing of large-scale graph data. Key features: - GQL图查询语言 - 点和边的属性索引 - Message passing接口 Tech stack: C++ Why it matters: Euler-2.0 simplifies the implementation of graph algorithms, making it easier to build and deploy distributed graph deep learning models on large-scale graph data. Repository: alibaba/euler GitHub stars: 2900 GitHub description: A distributed graph deep learning framework.

Overview: Agent Skills is a collection of instructions, scripts, and resources designed to help developers integrate AI agents with Supabase, a platform for building web applications using Supabase's suite of products and integrations. Key features: - Comprehensive Supabase development skill covering all Supabase products and integrations. - Postgres performance optimization guidelines. - Troubleshooting auth issues. - Using client libraries and SSR integrations. - Working with schema changes, migrations, and Postgres extensions. - Implementing indexes and query optimization. - Configuring connection pooling and scaling. - Reviewing database performance issues. - Working with Row-Level Security (RLS). - Helping set up Supabase Auth with Next.js. - Adding proper indexes to tables. Tech stack: TypeScript Why it matters: Agent Skills provides a structured way for developers to integrate AI agents with Supabase, making it easier to build and maintain applications with modern tools and practices.

### Overview: A fast terminal diff viewer and code review TUI, written in Rust, simplifies Git diff viewing and review. ### Key Features: - Side-by-side diff viewer with tree-sitter syntax highlighting - Review GitHub Pull Requests with `lumen diff --pr 123` - Annotate selections, hunks, or whole files - Watch mode and stacked-commit review - Optional AI commit messages and change explanations (10+ providers) - Works with Git and Jujutsu (jj) ### Tech Stack: Rust ### Why it Matters: - Simplifies Git diff viewing and review in the terminal. - Enhances collaboration and code review workflows. - Provides a fast and efficient way to review large diffs without leaving the terminal. ### Repository: jnsahaj/lumen ### GitHub Stars: 2899 ### Table of Contents: - Getting Started - Prerequisites - Installation - Usage - Visual Diff Viewer - AI Features - Coding Agent Integrations - Advanced Configuration ### Getting Started 🔅 ### Prerequisites Before you begin, ensure you have: - `git` installed on your system - [fzf](https://github.com/junegunn/fzf) (optional) - Required for `lumen explain --list` command - [mdcat](https://github.com/swsnr/mdcat) (optional) - Required for pretty output formatting ### Installation #### Using Homebrew (MacOS and Linux) ```bash brew

Overview: TensorRT_Pro is a C++ library based on TensorRT integration, designed to simplify the implementation of custom plugins and facilitate the deployment of high-performance inference with C++/Python support. Key features: - Simplified implementation of custom plugins - Encapsulated serialization and deserialization for easier usage - Facilitates the compile of fp32, fp16, and int8 models for deployment in server or embedded devices - Models compatible with RetinaFace, Scrfd, YoloV5, YoloX, Arcface, AlphaPose, CenterNet, and DeepSORT Tech stack: C++ Why it matters: TensorRT_Pro simplifies the implementation of custom plugins, making it easier to deploy high-performance inference with C++/Python support, and facilitates the compilation of models in various formats for deployment in different environments.

### Overview: This repository features a curated list of the latest breakthroughs in AI and data science, organized by release date. Each paper includes a clear video explanation, a link to a more in-depth article, and code (if applicable). The goal is to provide a comprehensive overview of the most interesting research in AI and data science, ensuring you don't miss any key papers. ### Key Features: - **Latest Breakthroughs**: Curated list of the latest AI and data science research papers. - **Clear Video Explanation**: Each paper comes with a detailed video explanation. - **Link to Article**: A link to a more in-depth article for further reading. - **Code**: Applicable papers include code for implementation and experimentation. ### Tech Stack: Python, Markdown, and GitHub Pages. ### Why it Matters: The repository is essential for staying updated on the latest advancements in AI and data science. It provides a clear, concise overview of the most interesting research, making it easier to understand and engage with the field. The inclusion of code allows for practical application and experimentation, making the research more accessible and impactful. ### Repository: [louisfb01/best_AI_papers_2021](https://github.com/louisfb01/best_AI_papers_2021) ### GitHub Stars: 2,896 ### Maintainer: [louisfb01](https://github.com/louisfb01) ### Subscribe to Newsletter: [louisbouchard.substack.com](https://louisbouchard.substack.com

Overview: Appllama is a design library that provides top-grossing app patterns, flows, and UI patterns with revenue and download context. These skills turn them into native-quality mobile screens. Key features: - Studying the best practices of top-grossing apps - Extracting design language and building screens that match - Using the Appllama MCP for native-feeling Expo / React Native screens - Apple HIG fidelity, semantic colors, native controls, anti-slop discipline, and a full-motion simulator loop Tech stack: Skills.sh Why it matters: These skills help agents build mobile apps that are as good as the best, with revenue and download context, and can be used with the Appllama MCP for a native-feeling design. Repository: Appllama/appllama-skills GitHub stars: 2895