Projects

### Overview: Bria RMBG 2.0 is a state-of-the-art background removal model designed to effectively separate foreground from background in various categories and image types, making it ideal for commercial use cases powering enterprise content creation at scale. The model has been trained on a carefully selected dataset, including general stock images, e-commerce, gaming, and advertising content, ensuring it meets the needs of content safety, legally licensed datasets, and bias mitigation. ### Key features: - **Background Removal Model:** Bria RMBG 2.0 is designed to remove backgrounds from images, enhancing visibility and clarity. - **Versatile Training Data:** The model is trained on a wide range of categories and image types, ensuring broad applicability. - **Commercial Ready:** This model is suitable for commercial use, with a commercial agreement available for purchase. ### Tech stack: Python ### Why it matters: Bria RMBG 2.0 is a significant improvement over previous versions, offering higher accuracy, efficiency, and versatility. It is ideal for use cases where content safety, legally licensed datasets, and bias mitigation are paramount. The model's non-binary output approach allows for flexible foreground-background separation, enhancing integration into complex pipelines. Repository: Bria-AI/RMBG-2.0 Primary language: Python GitHub stars: 345 GitHub description: Bria RMBG 2.0 - image background removal README excerpt: # BRIA Background Removal v2.0 <p align="center"><img src="https://platform.b

Overview: T2I-CompBench++ is a comprehensive benchmark designed for evaluating compositional text-to-image generation, showcasing its effectiveness in various scenarios and metrics. Key features: - 📝 A detailed benchmark for compositional text-to-image generation, covering multiple evaluation metrics. - 📈 Evaluation metrics adopted by Stable Diffusion 3, DALL-E 3, and PixArt-α, ensuring high accuracy and consistency. - 📈 Human evaluation of image-score pairs, providing a qualitative assessment of the generated images. - 🚀 Release of the dataset, training, and evaluation code, making it accessible for researchers and developers alike. Tech stack: Python Why it matters: T2I-CompBench++ serves as a crucial tool for benchmarking compositional text-to-image generation, offering a robust and consistent evaluation framework that enhances the quality and reliability of compositional AI models.

Overview: 3D-Adapter is an official PyTorch implementation of the papers "3D-Adapter: Geometry-Consistent Multi-View Diffusion for High-Quality 3D Generation" and "Generic 3D Diffusion Adapter Using Controlled Multi-View Editing". It aims to generate high-quality 3D models using geometry-consistent multi-view diffusion. Key features: - Geometry-Consistent Multi-View Diffusion - High-Quality 3D Generation Tech stack: JavaScript Why it matters: This project is essential for anyone working on 3D generation and rendering, as it provides a robust and efficient method for generating high-quality 3D models from multiple views. Repository: Lakonik/MVEdit Primary language: JavaScript GitHub stars: 347 GitHub description: 3D-Adapter: Geometry-Consistent Multi-View Diffusion for High-Quality 3D Generation

Overview: Diffusion4D is an official implementation of the NeurIPS 2024 paper "Diffusion4D: Fast Spatial-temporal Consistent 4D Generation via Video Diffusion Models". It provides a comprehensive dataset and rendering script for generating 4D videos from both image and text inputs. Key features: - Image-to-4D conversion - Text-to-4D conversion - 3D-to-4D conversion Tech stack: Python Why it matters: Diffusion4D offers a fast and efficient method for generating 4D videos, making it valuable for applications requiring real-time or near-real-time 4D video generation, such as simulations, training, and animation. Repository: VITA-Group/Diffusion4D Primary language: Python GitHub stars: 347 GitHub description: [NeurIPS 2024] Diffusion4D: Fast Spatial-temporal Consistent 4D Generation via Video Diffusion Models README excerpt: # Diffusion4D: Fast Spatial-temporal Consistent 4D Generation via Video Diffusion Models The official implementation of work "Diffusion4D: Fast Spatial-temporal Consistent 4D Generation via Video Diffusion Models". [[Project Page]](https://vita-group.github.io/Diffusion4D/) | [[Arxiv]](https://arxiv.org/abs/2405.16645) | [[Video (Youtube)]](https://www.youtube.com/watch?v=XJT-cMt_x

### Overview: MLM Scoring is a Python library designed to score sentences using masked language models like BERT, RoBERTa, and XLM. It provides various models and interfaces for both MXNet and PyTorch, making it versatile for different applications. ### Key Features: - **Scoring**: Scores sentences using pseudo-log-likelihood scores, log-probability scores, and maskless pseudo-log-likelihood scores. - **Rescoring**: Rescores n-best lists using log-linear interpolation. - **Supported Models**: BERT, RoBERTa, XLM, ALBERT, DistilBERT, GPT-2. - **Interfaces**: MXNet and PyTorch interfaces are available. - **Examples**: Demonstrates use cases such as speech recognition, machine translation, and linguistic acceptability. ### Tech Stack: - **Python**: The primary language used for development. - **MXNet**: Used for MXNet-based models. - **PyTorch**: Used for PyTorch-based models. ### Why It Matters: MLM Scoring is essential for evaluating the quality of generated text, improving speech recognition, and enhancing machine translation models. It provides a standardized way to measure and compare the performance of different language models. ### Repository: awslabs/mlm-scoring ### Primary Language: Python ### GitHub Stars: 347 ### GitHub Description: Python library & examples for Masked Language Model Scoring (ACL 2020) ### README Excerpt: This package uses masked LMs like [BERT](https://arxiv.org/

### Overview: Meissonic is an official implementation of the Meissonic model, a masked generative transformer designed for efficient high-resolution text-to-image synthesis. This project aims to revive and enhance the capabilities of masked generative transformers, making them more efficient and effective for generating high-resolution images from text inputs. ### Key Features: - **Efficient Text-to-Image Synthesis**: Meissonic uses masked generative transformers to synthesize high-resolution images from textual descriptions. - **Versatile Model Architecture**: The model is designed to handle various image resolutions, making it suitable for a wide range of applications. - **Comprehensive Documentation**: The project includes detailed documentation, including code examples and tutorials, to help developers integrate Meissonic into their projects. ### Tech Stack: - **Python**: The primary programming language used for the implementation. - **Masked Generative Transformers**: The backbone of the model, designed to generate high-resolution images from textual inputs. ### Why It Matters: Meissonic represents a significant advancement in the field of masked generative transformers, offering improved efficiency and performance for high-resolution text-to-image synthesis. Its implementation and documentation make it an excellent resource for developers looking to integrate this technology into their projects.

### Overview: This repository contains training, evaluation, and visualization code for our recent paper exploring loss-conditional diffusion models of neural network parameters. The core of G.pt is a transformer model that operates over sequences of parameters from the input neural network, leveraging very few domain-specific inductive biases. The model is trained as a diffusion model directly in parameter space and can optimize neural networks from random initialization in one step by prompting for a small loss/error or high return. ### Key features: - Training, evaluation, and visualization code for G.pt models - Five pre-trained G.pt DDPM Transformers for vision and RL tasks - Over 23M neural net checkpoints across 100K+ training runs - Dataset containing useful task-specific metadata for classification and reinforcement learning ### Tech stack: Python ### Why it matters: This repository provides a practical implementation of the G.pt model, showcasing its effectiveness in optimizing neural networks from random initialization. The pre-trained models and extensive dataset make it a valuable resource for researchers and practitioners in the field of neural network optimization.

### Overview: LLMVault is an intentionally vulnerable OWASP LLM Top 10 training platform designed for AI security training. It offers both Play Mode and Live Mode, allowing users to practice and learn from real-world attacks and defenses. ### Key Features: - **Play Mode**: 25 labs across three tiers (ten core, ten advanced, five expert) with scripted attacks and defenses. - **Live Mode**: Points at a real model running on the user's machine, with no scripted attacks or flags to look up. - **Dashboard**: Side-by-side comparison of Play and Live modes, with progress, rank, score, and live activity feed. - **Milestone Cards**: Pop up on badges earned in Play Mode, and notifications for Play and Live modes. - **Flag Verification**: Flags are verified by hash comparison instead of plaintext matches. ### Tech Stack: - Python 3.10+ - Docker - OWASP LLM Top 10 training ### Why It Matters: - Ideal for hands-on security education, focusing on OWASP LLM Top 10 vulnerabilities. - Helps developers and security professionals understand and mitigate AI security threats. ### GitHub Stars: 344 ### GitHub Description: An intentionally vulnerable OWASP LLM Top 10 training platform for AI Security, Prompt Injection, RAG Security, Agent Security, and GenAI penetration testing.

### Overview: 本项目是一个收集和整理机器学习系统（MLSys）领域优秀资源的网页，旨在为开发者提供高质量的学习和参考资料。资源包括来自学术界、业界专家和同龄人的推荐，涵盖了系统设计、算法实现和实践应用等多个方面。 ### Key features: - 收集来自学术界、业界专家和同龄人的MLSys资源推荐。 - 系统方向包括：方佳瑞、手抓饼熊、InterestingLSY、Yinmin Zhong、赵轩磊、谁来救救我、Fazzie、Ying Sheng、Lianmin Zheng、Chayenne Zhao、Hsword、Byron Hsu、Pentium PRO、ZZYSonny、刀刀宁、DefTruth、陈乐群、Zihao Ye、陈天奇、BBuf、Yineng Zhang、ZOMI酱、尘伊光、Kan Zhu、乔枫惜、KuangjuX、SiriusNEO、SuSun、拜月神使曌鹓鶵、进击的Killua、ZHANG Mingxing。 ### Tech stack: HTML ### Why it matters: 本项目为机器学习系统（MLSys）领域的开发者提供了宝贵的资源和学习资料，帮助他们快速入门和深入理解该领域的重要概念和实践方法。通过分享这些资源，我们希望能够促进MLSys领域的知识交流和技术创新。

### Overview: A live chatbot system designed for a final year project, built using Python and Natural Language Toolkit (NLTK) for Machine Learning. Easy to understand and implement, the chatbot is capable of simulating human-like conversations through text messages. ### Key Features: - **Python Implementation**: The chatbot is developed using Python programming language. - **Natural Language Toolkit (NLTK)**: Utilizes NLTK for Natural Language Processing (NLP) to enhance the chatbot's ability to understand and respond to user queries. - **Machine Learning**: The chatbot employs Machine Learning techniques for improved interaction and response accuracy. ### Tech Stack: - Python - NLTK - Machine Learning ### Why It Matters: A chatbot is a powerful tool for interaction between humans and machines, offering advanced capabilities in natural language processing. By using Machine Learning and NLTK, this project showcases the evolution of AI in practical applications, making it an essential project for final year students. ### Repository: The project is hosted on GitHub under the repository **Vatshayan/Live-Chatbot-for-Final-Year-Project**. It has received 347 GitHub stars, indicating its popularity and usefulness. ### Primary Language: The main language used for this project is **Jupyter Notebook**. ### GitHub Description: This chatbot system is designed for a final year project, utilizing Python and NLTK for Machine Learning. It is easy to understand and implement, making it a valuable tool for students and professionals in the field of artificial intelligence and natural language processing. ### Live Chatbot

### Overview: This React Native plugin provides a seamless way to integrate Apple's Intelligence Foundation Models into your mobile applications, allowing for on-device processing and privacy-focused AI capabilities. ### Key features: - **On-device Apple Intelligence**: Access Apple's Foundation Models locally. - **Privacy-focused**: All processing happens on-device, no data sent to servers. - **Structured JSON output**: Generate structured data with JSON schemas. - **Text generation**: Create human-like text responses. - **Session management**: Configure and manage LLM sessions. - **TypeScript support**: Full type safety and IntelliSense. - **Custom tools**: User-defined tool-box available for LLM use. ### Tech stack: Swift ### Why it matters: This plugin is essential for building AI-powered mobile apps with privacy-first approaches, chatbots, educational tools, and more. It offers a privacy-focused solution that can be used in various applications, making it a valuable addition to any developer's toolkit.

### Overview: Meilisearch .NET is the Meilisearch API client for C# developers. ### Key features: - **Indexes**: Store and manage your documents. - **Documents**: Create and update documents in your index. - **Get Task Information**: Retrieve task status and details. - **Search**: Perform search queries on your indexed documents. ### Tech stack: C# ### Why it matters: It simplifies integrating Meilisearch with C# applications, making it easier to build and deploy search functionalities.

Overview: This repository contains the official implementation of the paper "Model-based Diffusion for Trajectory Optimization". It provides a novel diffusion-based trajectory optimization framework that utilizes a dynamics model to generate high-quality trajectories. Key features: - Model-based diffusion (MBD) framework - Supports multiple environments including `hopper`, `halfcheetah`, `walker2d`, `ant`, `humanoidrun`, `humanoidstandup`, `humanoidtrack`, `car2d`, and `pushT` - Includes both model-based diffusion and model-based diffusion combined with demonstrations - Provides scripts for running multiple seeds, visualizing the diffusion process, and black-box optimization - Built on top of Brax for environment and RL implementation Tech stack: Python Why it matters: MBD outperforms existing methods in terms of sample efficiency and generalization, making it a significant advancement in trajectory optimization. Repository: LeCAR-Lab/model-based-diffusion GitHub stars: 345

Overview: This repository is a curated collection of all the "Awesome --Topic Name--" GitHub repositories that I have found relevant to Data lifecycle, Machine Learning, Deep Learning, and distributed computing. Key features: - Data augmentation - Multitask learning - Diffusion models - Self-supervised learning - Semi-supervised learning - Weakly supervised learning - Learning with label noise - Adversarial ML/DL - Architecture search - Contrastive self-supervised learning - Zero-shot learning - Few-shot learning - Siamese networks - Image classification - Contrastive learning - Visual transformers - Transformers in medical imaging - Transformers Tech stack: Python, JavaScript, Markdown, and various open-source tools. Why it matters: This repository serves as a comprehensive resource for individuals looking to start in the topics of Data lifecycle, Machine Learning, Deep Learning, and distributed computing. It provides a curated list of repositories that are believed to be important for beginners and serves as a starting point for further exploration and learning.

Overview: Bili Note 是一个基于 B 站视频内容的 Markdown 笔记工具，旨在帮助用户整理和学习 B 站上的知识和教程。它通过完整归档字幕、图文、图片和评论，以及动态长度控制，提供高质量的学习笔记。 Key features: - 完整归档：保存字幕、图文、图片和评论，支持多模态输入。 - 动态长度：根据内容信息量和结构调整笔记长度。 - 质量感知：结合热度、互动和发布时间调整笔记预算。 - 写前定标：生成推荐字数和写作粒度，写后评分用于验收。 - 字幕密度提醒：识别稀疏字幕并提示关键帧。 - 关键帧视觉理解：使用关键帧生成视觉联系图。 Tech stack: Python Why it matters: Bili Note 提供了一个系统的方法来整理和学习 B 站上的知识，帮助用户在学习过程中保持高质量和可追溯性。 Repository: Rimagination/bili-note Primary language: Python GitHub stars: 343 GitHub description: Extract Bilibili videos into learning-oriented Markdown notes with full subtitle and comment archives

### Overview: Photoswap is an innovative method for personalizing subject swapping in images, introduced at NeurIPS 2023. This project provides an official implementation of the NeurIPS 2023 paper "Photoswap: Personalized Subject Swapping in Images." ### Key Features: - **Model Architecture**: Utilizes a diffusion model with concept learning capabilities. - **Training**: Learns new subjects as new tokens, improving concept accuracy through diverse training data. - **Concept Learning**: Supports training with Text Inversion, DreamBooth, and custom diffusion models. - **Attention Swap**: Enables subject swapping by adjusting the swapping step and text prompt for optimal results. ### Tech Stack: Jupyter Notebook ### Why It Matters: Photoswap offers a novel approach to personalized image editing, making it a significant advancement in the field of image manipulation and AI.

### Overview: This project is a ChatGPT-based prompt generation model designed for MidJourney, aiming to simplify image creation and enhance creativity through partial hints. ### Key Features: - **Model Integration**: Uses ChatGPT to generate prompts for MidJourney. - **Partial Hints**: Introduces partial hints to stimulate creativity and provide new ideas. - **CLI and Discord Bot**: Available soon, providing command-line interface and Discord integration. ### Tech Stack: Python ### Why it Matters: This model simplifies the creation of images while enhancing creativity through partial hints, making it a valuable tool for artists and designers. ### Repository: awekrx/ChatGPT-MidJourney-prompt ### GitHub Stars: 345 ### Installation: ```bash pip install chatGPTMidJourneyPrompt ``` ### Usage: ```py from chatGPTMidJourneyPrompt.mjPrompt import PromptGenerator # Supported authorization methods: via email and password, via token, via API key config = { "email": "your_email", "password": "your_password", # or "session_token": "your_session_token", # or "api_key": "your_api_key", } promptGenerator = PromptGenerator(config) prompt = promptGenerator.V5("any text") prompt = promptGenerator.V4("any text") prompt = promptGenerator.niji("any text") prompt = promptGenerator.testp("any text") # or advanced usage if needed promptConfig = { "model": "artistic

### Overview: This repository contains a curated list of awesome papers focused on **Semantic Retrieval**, including early methods and recent neural models for information retrieval tasks such as ad-hoc retrieval, open-domain QA, community-based QA, and automatic conversation. ### Key Features: - **Survey Paper**: A comprehensive review of semantic models for the first-stage retrieval. - **Chapter 1: Classical Term-based Retrieval**: Covers vector space models, term weighting approaches, and relevance weighting. - **Chapter 2: Early Methods for Semantic Retrieval**: Includes query expansion, document expansion, term dependency models, and topic models. - **Chapter 3: Neural Methods for Semantic Retrieval**: Discusses sparse and dense retrieval methods, as well as hybrid retrieval techniques. - **Chapter 4: Other Resources**: Covers other tasks, datasets, and indexing methods. ### Tech Stack: - **Programming Languages**: Unknown ### Why It Matters: This repository serves as a valuable resource for researchers and developers interested in semantic retrieval models, providing a comprehensive overview of both classical and modern approaches to information retrieval tasks.

### Overview: Syftr is an agent optimizer designed to help developers find the best agentic workflows for their budget. Leveraging advanced multi-objective Bayesian Optimization and a novel Pareto Pruner, Syftr efficiently samples a search space to estimate a Pareto-frontier balancing accuracy and competing objectives. ### Key Features: - **Pareto-Optimal Flows**: Finds the best combination of parameters for a given budget. - **Multi-Objective Bayesian Optimization**: Utilizes advanced algorithms for efficient parameter tuning. - **Pareto Pruner**: Optimizes the search space for agentic and non-agentic flows. - **Flexible Dataset Interface**: Supports datasets from HuggingFace Datasets for seamless integration. ### Tech Stack: - **Ray**: Distributed and scalable search engine for large clusters. - **Optuna**: Flexible define-by-run interface and support for multi-objective optimization. - **LlamaIndex**: Builds sophisticated agentic and non-agentic RAG workflows. - **HuggingFace Datasets**: Fast, collaborative, and uniform dataset interface. - **Trace**: Optimizes textual components within workflows. ### Why It Matters: Syftr is a powerful tool for developers looking to optimize AI workflows. By efficiently sampling a search space and estimating Pareto-frontiers, Syftr helps find the best balance between accuracy and competing objectives, making it a valuable asset in the field of agent optimization.

Overview: displaCy.js is an open-source NLP visualiser designed to help developers compare different services and explore their own in-house models. It integrates seamlessly with spaCy's syntactic parser, making it a valuable tool for developers working with statistical models. Key features: - Integrates seamlessly with spaCy's syntactic parser - Provides a modern and service-independent visualisation library - Facilitates easy comparison of different services and in-house models - Enables sharing of visualisations for discussion and further use - Supports running the service yourself for development purposes Tech stack: JavaScript Why it matters: displaCy.js simplifies the process of visualising and comparing NLP models, making it easier for developers to evaluate and improve their text analysis tools.