Projects

### Overview: Tokenize Anything via Prompting is a unified model designed to segment, recognize, and caption arbitrary regions with flexible visual prompts. It leverages extensive segmentation masks from SA-1B and pre-trained EVA-CLIP for improved performance. ### Key Features: - **Flexible Prompting**: Supports point, box, and sketch prompts. - **Unified Model**: Seamlessly integrates segmentation, recognition, and captioning. - **Exhaustive Training**: Trained with SA-1B masks and EVA-CLIP for high accuracy. ### Tech Stack: - Jupyter Notebook ### Why It Matters: This model is pivotal in enabling versatile and effective visual language processing, making it essential for applications requiring precise and context-aware text generation from images.

### Overview: Sidecar is the AI brains for Aide editor, responsible for integrating and managing AI tools locally on your machine. ### Key features: - Tool box for all AI tools - Symbol folder for independent code handling - Prompts for inline completions - Repository map for code symbol management ### Tech stack: Rust ### Why it matters: Sidecar enhances the functionality of Aide editor by integrating AI tools, ensuring smooth collaboration and productivity. ### Repository: codestoryai/sidecar ### GitHub stars: 600 ### GitHub description: Sidecar is the AI brains for Aide editor and works alongside it, locally on your machine.

### Overview: This repository contains Docker skills tailored for AI coding agents, designed to help them build, test, debug, and optimize containerized apps with consistent, reusable workflows. ### Key Features: - **Dockerfile & Build**: Guidance for initializing and structuring a Dockerized project. - **Docker Compose**: Strategies for efficient, secure, and optimized Docker image builds. - **Docker Sandboxes**: Patterns for robust, maintainable Docker Compose configurations. - **Docker Agent**: Rules for authoring, running, and distributing AI agents with Docker Agent. ### Tech Stack: Python ### Why it Matters: These skills provide a structured approach to Docker-related tasks, ensuring consistency and reusability across different environments.

Overview: MeTRAbs is an absolute 3D human pose estimator that utilizes metric-scale truncation-robust heatmaps for estimating human poses from RGB images. The repository includes code for the paper "MeTRAbs: Metric-Scale Truncation-Robust Heatmaps for Absolute 3D Human Pose Estimation" by István Sárándi, Timm Linder, Kai O. Arras, and Bastian Leibe. Key features: - Improved models based on EfficientNetV2 and MobileNetV3 - Support for multiple skeleton conventions - Simplified API - Radial/tangential distortion handling - Plausibility filtering - Full codebase migrated to TensorFlow 2 and Keras - Released best-performing models based on EfficientNetV2 and super fast ones using MobileNetV3 Tech stack: Python Why it matters: MeTRAbs is a significant advancement in 3D human pose estimation, offering robustness against metric-scale changes and improved performance through efficient models and simplified APIs.

Overview: Point Transformer is a self-attention layer implementation for Pytorch that significantly improves point cloud classification and segmentation. Key features: - Supports Pytorch framework - Implements the Point Transformer architecture - Includes a simple circuit for better performance Tech stack: Pytorch Why it matters: The implementation allows researchers to outperform previous methods in point cloud classification and segmentation tasks, showcasing the potential of the Point Transformer layer.

### Overview: This repository is a curated list of resources for learning about natural language processing (NLP) and text analytics. It includes books, blogs, articles, case studies, and more to help you understand and apply NLP techniques effectively. ### Key Features: - **Books:** - Text Mining with R - Mastering Text Mining with R - Text Mining in Practice with R - Natural Language Processing with PyTorch - Applied Text Analysis with Python - **Blogs:** - General articles - Biases in NLP - Scraping - Cleaning - Stemming - Dimensionality Reduction - Sarcasm Detection - Document Classification - Entity and Information Extraction - Document Clustering and Document Similarity - Concept Analysis/Topic Modeling - Sentiment Analysis - Text Summarization - Machine Translation - Q&A Systems, Chatbots - Fuzzy Matching, Probabilistic Matching, Record Linkage - Word and Document Embeddings - Transformers and Language Models - Deep Learning - Knowledge Graphs - **Major NLP Conferences:** - NAACL (Natural Language Processing Conference) - ACL (Association for Computational Linguistics) - EMNLP (Conference on Empirical Methods in Natural Language Processing) - **Benchmarks:** - Various benchmarks for evaluating NLP models - **Online Courses:** - Introduction to NLP with

Overview: NumPyCNN is a Python implementation for building Convolutional Neural Networks (CNNs) from scratch using NumPy. It focuses on the forward pass of training CNNs without learning algorithms. Key features: - Implementing all classes and functions needed to build a CNN - No learning algorithm used, just a learning rate for weight changes after each epoch Tech stack: Python Why it matters: This project is useful for developers looking to understand the basics of CNNs and implement them from scratch using NumPy, which is a fundamental skill in deep learning.

Overview: `chat-bubble` is a simple chatbot UI for the web that allows developers to create interactive conversations using JSON scripting. Key features: - Quick set-up and implementation - Works with or without Natural Language Classifiers - 1KB GZipped, no dependencies, written with ES5 (compatible with IE11+) Tech stack: JavaScript Why it matters: It provides a user-friendly interface for developers to create engaging chatbot experiences without the need for complex server-side scripting, making it ideal for quick prototyping and small-scale projects.

### Overview: EMILIA Protocol provides an authority control plane for autonomous work, ensuring that payments and transactions are authorized and verifiable. It leverages open protocols to maintain evidence's independence and checkability. ### Key features: - **Open Protocol:** Keeps evidence verifiable and independently checkable under the verifier's trusted keys and rules. - **Gate Enforcement:** Ensures that the payment or transaction is authorized before it is executed, preventing unauthorized changes. - **Agent Preparation:** Prepares and manages payments, ensuring that the agent can only release the payment after being approved by the system. ### Tech stack: TypeScript ### Why it matters: EMILIA Protocol is essential for maintaining the integrity and accountability of financial transactions and payments, ensuring that the evidence of such transactions is verifiable and traceable. It provides a secure and transparent way to control and manage payments, making it crucial for businesses and organizations that need to ensure the authenticity and reliability of their financial processes. Repository: emiliaprotocol/emilia-protocol Primary language: TypeScript GitHub stars: 598 GitHub description: Authority control plane for autonomous work. EMILIA Gate enforces finite customer-owned mandates at protected executor boundaries; the open protocol keeps evidence verifiable.

Overview: This repository contains a TensorFlow implementation of a text-to-image synthesis model based on the GAN-CLS algorithm from the paper "Generative Adversarial Text-to-Image Synthesis". Key features: - Text-to-image synthesis using GAN-CLS algorithm - Built on TensorLayer framework - Supports training on the Oxford-102 flower dataset - Captions for images are extracted from the paper Tech stack: Python, TensorFlow, TensorLayer, NLTK Why it matters: This model can generate realistic images from textual descriptions, which has applications in various fields such as virtual reality, content generation, and more. Repository: zsdonghao/text-to-image GitHub stars: 599 GitHub description: Generative Adversarial Text to Image Synthesis / Please Star

Overview: LAMBDA is a data analysis agent designed to turn natural-language questions into reproducible analysis workflows. It integrates Python, Shell, and FastAPI for conversational data analysis, providing a persistent workspace, automatic chart and report generation, and supports Chinese and English interfaces. Key features: - Conversational data analysis with executable Python and shell tools. - Beautiful and intuitive UI for data analysis. - Autonomous dataset exploration for CSV, Excel, text, and other common files. - Persistent workspace per conversation, keeping variables and generated files available. - Automatic chart, table, report, notebook, and artifact tracking in the Files panel. - Export options for Jupyter Notebooks, Markdown reports, PDF reports, and slides. - Support for OpenAI-style model endpoints and model lists. - Local SQLite storage for conversations, uploads, and generated artifacts. Tech stack: Python, Node.js, FastAPI, React, TypeScript, SQLite, OpenAI-compatible models. Why it matters: LAMBDA simplifies complex data analysis tasks, making it accessible and efficient for developers and researchers.

### Overview: This repository provides a comprehensive guide to Natural Language Processing (NLP) through 10 kernels, covering tokenization, preprocessing, bag of words and similarity, TF-IDF and document search, Naive Bayes text classification, LDA topic modeling, word embeddings, recurrent neural networks (RNNs) and language modeling, machine translation with attention, and transformers. The primary language used is Jupyter Notebook, and it is designed to be accessible and easy to follow, starting from basic concepts and gradually building up to advanced topics. ### Key Features: - Tokenization - Preprocessing - Bag of Words and Similarity - TF-IDF and Document Search - Naive Bayes Text Classification - LDA Topic Modelling - Word Embeddings - Recurrent Neural Networks (RNNs) and Language Modelling - Machine Translation and Attention - Transformers ### Tech Stack: Jupyter Notebook ### Why it Matters: This repository serves as a practical guide for beginners to understand the core concepts of NLP, from basic tokenization to advanced topics like transformers. It provides a step-by-step approach to building models and understanding how they work, making it ideal for developers looking to quickly grasp the fundamentals of NLP. Repository: JUSTSUJAY/nlp-zero-to-hero

Overview: This repository contains a corpus of articles scraped from WeChat public accounts, cleaned of HTML, and formatted in JSON. Each entry includes the account name, ID, title, and content. Key features: - JSON formatted articles - Cleaned from HTML - Compressed in zip files without password - Available for research use only Tech stack: Python, JSON, HTML parsing Why it matters: This corpus is useful for researchers and developers looking to study WeChat public account content, providing a rich dataset for analysis and language processing tasks. Repository: nonamestreet/weixin_public_corpus GitHub stars: 598 GitHub description: 微信公众号语料库

Overview: goclaw is an open-source AI assistant framework built with Go, designed to provide a powerful AI agent framework. Key features: - A complete tool system including FileSystem, Shell, Web, Browser, Docker sandbox, and permission control. - Skills system that supports OpenClaw and AgentSkills specifications, enabling automatic discovery and environment准入 control (Gating). - Persistent session storage using JSONL, with a record and recovery of tool calls. - Support for multiple channels including Telegram, WhatsApp, Feishu, QQ, WeWork, DingTalk, Infoflow, Gotify, Slack, Discord, Microsoft Teams, WeChat, and Google Chat. - Flexible configuration support for YAML/JSON, hot reloading, and environment variables. - Multi-LLM providers support including OpenAI, Qianfan (compatible with OpenAI), Anthropic, and OpenRouter, with fault-tolerance. - WebSocket gateway for real-time communication. - Built-in cron scheduler. - Browser automation using Chrome DevTools Protocol. - Memory system for vector database and QMD (Quick Markdown Database). - Multi-account support per channel. - Cross-platform support for Linux, macOS, and Windows. Tech stack: Go Why it matters: goclaw offers a comprehensive AI assistant framework built on Go, providing a powerful and flexible platform for developers to build their own AI agents. GitHub stars: 598

Overview: Spine Animation AI is an AI-powered Claude skill designed to automate the creation, rigging, and refinement of Spine 2D skeletal animations. It includes auto-positioning of body parts, skeleton JSON generation, animation generation, interactive HTML5 previews, and correction and refinement of existing skeletons. Key features: - Auto-position body parts from reference images using SIFT + RANSAC - Build skeleton JSON with proper bone hierarchy and draw order - Generate animations (idle, walk, run, attack, wave, jump) using the 12 principles of animation - Produce interactive HTML5 previews using the official Spine Web Player - Correct and refine existing skeletons with AI-assisted offset adjustments Tech stack: Python Why it matters: This skill streamlines the tedious parts of Spine 2D animation production, allowing artists to focus on creative aspects while AI handles the technical tasks, leading to more efficient and accurate animations. Repository: GenielabsOpenSource/spine-animation-ai Primary language: Python GitHub stars: 585 GitHub description: AI-powered Claude skill for Spine 2D skeletal animation — auto-rig, animate, and preview characters README excerpt: An AI-powered Claude skill for creating, rigging, and refining Spine 2D skeletal animations — from raw assets to interactive previews. The skill includes auto-positioning, skeleton JSON generation, animation generation, interactive HTML5 previews, and correction and refinement of existing skeletons. It works with Claude projects or custom skill directories.

### Overview: This repository provides an analysis-by-synthesis framework for fitting a textured FLAME model to in-the-wild images. FLAME is a lightweight generic 3D head model learned from over 33,000 head scans, but it does not come with an appearance space. This repository describes how to build a texture space for FLAME from in-the-wild images, provides code to fit a textured FLAME model to images, and optimizes for FLAME's parameters, appearance, and lighting. The FLAME model and the texture space can be downloaded from the FLAME project website, where you need to sign up and agree to the license for access. ### Key Features: - **Texture Space Construction**: Builds a texture space from in-the-wild images to cover a large range of ethnicities, age groups, etc. - **Model Fitting**: Fits FLAME to in-the-wild images, optimizing for FLAME's parameters, appearance, and lighting, and optimizing for the FLAME texture to match an in-the-wild image. - **Code Implementation**: Provides code to build a texture space, fit a textured FLAME model, and optimize for the FLAME texture to match an in-the-wild image. ### Tech Stack: Python ### Why It Matters: This repository provides a comprehensive solution for creating a texture space for FLAME, enabling the fitting of FLAME models to in-the-wild images. The use of FLAME and the texture space can be beneficial in various applications, such as creating realistic textures for virtual avatars or enhancing the appearance

Overview: This repository contains a curated list of resources and advice useful for a PhD in computer vision, compiled by the author. Key features: - General advice on conducting research - Better/faster paper reading tools and resources - Tips for writing a good CVPR/ECCV/ICCV paper - How to write a good review - How to release code that is easy to understand and reuse - Tools for fast and reproducible experimentation (in Python/PyTorch) - Tricks for neural networks - Some coding tips Tech stack: Python/PyTorch Why it matters: This repository serves as a concise and accessible resource for researchers in computer vision, providing a quick overview of essential tools and advice. It helps researchers efficiently manage their research and stay updated with the latest practices in the field.

### Overview: Agent-Me is an open-source, inspectable AI Agent Twin that distills your knowledge, memories, and decisions into an AI representation of yourself. It provides a stable, version-controlled knowledge base and a reviewable memory substrate, allowing for durable identity and memory models. ### Key Features: - **Reviewable Knowledge Base**: Uses Markdown for version-controlled knowledge, providing a stable and inspectable body of personal information. - **Memory Substrate**: Supports SQLite-backed typed memory, including reviewed identity/relationships, declared valid and event times, and persisted chat transcripts. - **Memory Workbench**: Offers a browser review workbench for approved-source ingestion, conflict review, provenance, and local evidence-grounded answers. - **Scoped Model-Assisted Extraction**: Proposes pending classifications over exact source quotations for more precise and contextually relevant responses. ### Tech Stack: - **Python**: The primary programming language used for the implementation. - **FastAPI**: For the backend API. - **React**: For the frontend user interface. ### Why It Matters: Agent-Me aims to bridge the gap between a personal AI and a fully realized, contextually aware representation of a person. By providing a stable, reviewable knowledge base and a system that can reason and verify its own responses, it opens the door to building more comprehensive and useful AI systems that can truly represent and interact with individuals in a meaningful way. ### Repository: - **Repository**: jzjzzzzzzz/agent-me - **GitHub Stars**: 586 - **Primary Language**: Python - **

Overview: Holodeck is a high-fidelity simulator designed for reinforcement learning and robotics research, built on Unreal Engine 4. Key features: - 7+ rich worlds for training agents in, and many scenarios for those worlds - Linux and Windows support - Easily extend and modify training scenarios - Train and control more than one agent at once - Simple, OpenAI Gym-like Python interface - High performance - simulation speeds of up to 2x real time are possible. Performance penalty only for what you need - Run headless or watch your agents learn Tech stack: Python Why it matters: Holodeck provides a powerful and flexible environment for researchers to test and refine their reinforcement learning algorithms and robotics models, offering a high-fidelity simulation that can be easily extended and modified to suit various research needs.

Overview: The SSBM "Phillip" AI is a project based on Deep Reinforcement Learning, designed as a successor to the original "Phillip" project. It is no longer actively maintained but has a successor project based on imitation learning. Key features: - Runs on various operating systems including Linux, macOS, and Windows. - Supports Dolphin emulator for SSBM. - Includes a set of pre-trained agents, with the best human-like agent being "delay18/FalcoBF". - Supports training and running on slurm clusters for advanced users. Tech stack: Python 3 Why it matters: The project provides a powerful AI assistant for SSBM players, offering a unique combination of AI and human-like playstyle.