Research & Education

Afriwise is an online platform designed to simplify African law and compliance for businesses. It provides instant answers to critical legal and business questions across various African countries, helping users find local counsel. The platform features Afriwise 360° Legal & Regulatory Intelligence, offering practical guidance, tailored comparison reports, and the ability to request clarifications from top law firms. Additionally, the Afriwise Laws & Monitoring solution provides fully maintained local legal frameworks, access to official texts with English translations, and automated email alerts for regulatory changes. It aims to reduce external legal costs and ensure businesses operate on the right side of the law across the continent.

Marbyt is a technical consultancy focused on developing specific software for the research world, particularly in biotechnology. As researchers themselves, they understand the complex and repetitive nature of laboratory work and leverage their informatics expertise to automate or simplify many tasks with algorithms. This approach aims to save significant time and money for investigators. Marbyt provides personalized solutions for research projects, emphasizing a biotechnological focus to ensure biological interpretation remains central. They offer bioinformatic tools tailored to facilitate daily research, reduce costs, and provide ongoing support and accompaniment throughout collaborations. Their services cater to sectors like agro-foodtech and healthcare, aiming to advance science and unlock research potential.

UnifierAI is an AI-based research and development company dedicated to fostering innovation through strong research competence and creative collaborations. The company's core mission is to advance technology by leveraging AI-driven research initiatives. While specific features and services are not detailed on their public website, their focus on R&D suggests they are involved in developing cutting-edge AI solutions and potentially offering research services or platforms to other organizations. Their emphasis on innovation and collaboration indicates a forward-thinking approach to AI development.

Recontact.ai is currently a parking page, indicating the domain is registered but not actively hosting the intended AI tool. The page is provided by Spaceship, a digital platform offering various services to establish an online presence. These services include domain name registration, shared hosting with high-performance servers and scalable plans, and encrypted branded email through Spacemail. Spaceship emphasizes ease of use, providing tools to help users find suitable domain names, launch websites without needing developers, and automatically connect domains with hosting plans. It also highlights security features like domain privacy and free SSL for one year with every website created.

CNNMRF is a Torch-based implementation for image synthesis, leveraging the power of Markov Random Fields and Convolutional Neural Networks. This tool is designed for both unguided image synthesis, such as classical texture generation, and guided image synthesis, which includes transferring styles between different images. Users can transform a photo into a painting using a reference style, or balance content and style in the resulting image. The project provides detailed setup instructions for Ubuntu with CUDA 10 and CUDNN 7.6.2, along with guidance on installing Torch and downloading pre-trained VGG networks. It offers command-line parameters for customizing style, content, and output image sizes, making it a flexible solution for researchers and developers interested in advanced image manipulation.

Magic UI offers a comprehensive collection of reusable UI components, blocks, and templates designed to help developers build visually appealing landing pages and marketing materials. Built with modern web technologies like React, TypeScript, Tailwind CSS, and Motion, it provides over 150 free and open-source animated components and effects. The platform also features Magic UI Pro, which includes 50+ blocks and templates for building production-ready landing pages in minutes, saving significant development time. It emphasizes good design philosophy, aiming to establish trust with users through polished and professional interfaces, drawing inspiration from tools like shadcn/ui.

OpenProtein.AI offers a sophisticated cloud platform for AI-driven protein design and optimization, leveraging advanced models like PoET-2, AlphaFold2, ESM2, and Clustal Omega. The platform helps users reduce costs, accelerate projects, and achieve better results in fewer rounds by eliminating guesswork and streamlining workflows. It allows for the design of optimized variant libraries, comparison of library outcomes, and visualization of predicted protein structures. OpenProtein.AI supports projects of all sizes, from 96-well plates to high-throughput pipelines, and can optimize any protein for various properties like activity, expressibility, and thermo stability. The technology learns from natural protein sequence databases and user-specific data to generate novel, functional proteins.

Cap is an open-source screen recording tool designed to be a lightweight, powerful, and cross-platform alternative to services like Loom. It enables users to quickly record their screens and share the content with ease. The tool focuses on providing a seamless experience for capturing visual information, making it suitable for various use cases from tutorials to quick demonstrations. Its open-source nature suggests a community-driven development and potentially greater transparency and customization options for users. Cap aims to simplify the process of creating and distributing screen recordings.

MIRNet is an advanced AI tool designed for real image restoration and enhancement, leveraging enriched features to deliver state-of-the-art results across various image processing tasks. Its novel architecture maintains spatially-precise high-resolution representations while integrating strong contextual information from low-resolution representations. The core of MIRNet is a multi-scale residual block that incorporates parallel multi-resolution convolution streams for feature extraction, information exchange across these streams, and spatial and channel attention mechanisms for capturing contextual details. This allows for attention-based multi-scale feature aggregation, combining contextual information from multiple scales while preserving high-resolution spatial details. MIRNet excels in image denoising, super-resolution, and image enhancement, as demonstrated by extensive experiments on five real image benchmark datasets.

DeepSurv implements a deep learning generalization of the Cox proportional hazards model using Theano and Lasagne. This approach offers an advantage over traditional Cox regression by adaptively learning covariates, removing the need for their a priori selection. The tool is versatile and can be applied in numerous survival analysis applications. A notable medical application, `recommend_treatment`, is provided, which offers treatment recommendations for patient observations. DeepSurv is open-source and provides clear instructions for installation, dependency management, running experiments with Docker, and training/evaluating networks, making it accessible for technical users in research and development.

motion-diffusion-model is an open-source PyTorch implementation of the "Human Motion Diffusion Model" paper, designed for generating and editing human motion sequences. The tool boasts significant speed improvements, now running 40X faster with a 50-diffusion-step model and optimized CLIP calling. It supports various tasks including text-to-motion, action-to-motion, and unconstrained motion synthesis. Users can generate motions from text prompts or actions, render SMPL meshes, and perform motion editing such as in-between and upper-body modifications. The project also integrates DiP for ultra-fast text-to-motion and offers features like DistilBERT text encoder support and dataset caching for faster loading.

DeepLabCut is an open-source toolbox designed for state-of-the-art markerless pose estimation across various animals and humans. It leverages deep learning to track user-defined features, making it highly versatile and applicable to a wide range of behaviors and species. The tool provides a user-friendly GUI and API, integrating advanced models and frameworks while offering sensible defaults for life scientists. It supports both single and multi-animal pose estimation, identification, and tracking. DeepLabCut is actively maintained, offering continuous improvements, including faster performance variants, real-time capabilities, and a recent backend migration to PyTorch for enhanced flexibility and easier installation. Comprehensive documentation, online courses, and a model zoo are available to assist users.

Finscale AI is an open-source platform dedicated to financial data residency, operating within the Open Constitution AI network. It functions as a research and development lab, providing AI processors to its network constituents. The platform is built upon the Open Constitution licensing framework, emphasizing the development of digital commons infrastructure. Finscale aims to support the secure and localized management of financial data, leveraging AI to enhance its capabilities for various financial applications and research within its ecosystem.

DiscoFaceGAN is a TensorFlow-based implementation for disentangled and controllable face image generation, as presented in a CVPR 2020 Oral paper. This tool allows for the creation of virtual people's faces with precise control over identity, expression, pose, and illumination. It achieves this through 3D imitative-contrastive learning, embedding 3D priors into adversarial learning to imitate the image formation of a 3D face deformation and rendering process. A key feature is its factor disentanglement, ensuring that changing one factor (e.g., expression) does not affect others. The tool also supports reference-based generation, real image pose manipulation, lighting editing, and expression transfer, making it valuable for researchers and developers working with facial image synthesis and manipulation.

OBBDetection is an open-source oriented object detection library built upon MMdetection v2.2, designed for researchers and developers working with object detection tasks. It inherits all features from MMdetection, ensuring a robust and familiar environment. The library supports multiple frameworks and implements various oriented object detectors like RoI Transformer and Gliding Vertex. A key feature is its flexible representation of oriented boxes, accommodating Horizontal Bounding Boxes (HBB), Oriented Bounding Boxes (OBB), and 4-point boxes (POLY). It leverages BboxToolkit for oriented bounding box operations and includes a model zoo with benchmarks for supported methods and backbones. The project is released under the Apache 2.0 license.

PET (Pattern-Exploiting Training) is an open-source research tool designed for few-shot text classification and natural language inference. It employs a semi-supervised training procedure that reformulates input examples as cloze-style phrases, allowing language models to better understand tasks. The tool, along with its iterative variant iPET, demonstrates significant performance improvements over traditional supervised training and other semi-supervised baselines, even surpassing GPT-3 in some low-resource scenarios while requiring substantially fewer parameters. It supports various training modes including PET, iPET, and supervised training, and offers evaluation methods like unsupervised and priming. Researchers can use PET for 13 different tasks, including SuperGLUE tasks, and can also customize it for their own specific applications by defining DataProcessors and PVPs (Pattern-Verbalizer Pairs).

Face-Pose-Net provides a DCNN model and Python code for robustly estimating 6 degrees of freedom (6DoF) 3D face pose or 11 parameters of a 3x4 projection matrix from unconstrained images. A key differentiator is its ability to perform face alignment without relying on fragile landmark detectors, making it highly effective even with low-resolution, occluded, or near-profile views. The tool integrates with a Face Renderer to create an end-to-end pipeline for facial pose estimation and generating multiple rendered views for alignment and data augmentation. It supports both CPU and GPU for extremely fast pose estimation and offers improved face recognition through better face alignment compared to state-of-the-art landmark detectors.

few-shot-object-detection (FsDet) offers official implementations of few-shot object detection benchmarks, including the ICML 2020 paper "Frustratingly Simple Few-Shot Object Detection." It introduces new benchmarks across PASCAL VOC, COCO, and LVIS datasets, with multiple groups of few-shot training examples and evaluation results for both base and novel classes. The repository provides benchmark results and pre-trained models for a two-stage fine-tuning approach (TFA), where the detector is first trained on abundant base classes and then fine-tuned on a small balanced training set. FsDet is modular, allowing for easy integration of custom datasets and models, serving as a general framework for future research in few-shot object detection.

PassGAN is an open-source deep learning tool for password guessing, implementing the approach described in the paper "PassGAN: A Deep Learning Approach for Password Guessing." This repository provides a modified TensorFlow implementation of Improved Training of Wasserstein GANs, making it easy to train and sample from the model. It includes a command-line interface for generating password samples and training custom models. A pretrained PassGAN model, trained on the RockYou dataset, is also provided. Users can train their own models using various password leaks and datasets, with instructions for downloading common datasets like the LinkedIn leak. The tool is released under an MIT License, acknowledging the original authors of the PassGAN paper and the underlying WGAN training code.

SparseR-CNN is an advanced end-to-end object detection model that leverages learnable proposals, eliminating the need for hand-crafted proposals common in traditional object detection systems. This approach allows for more efficient and potentially higher-performing detection across various computer vision applications. The tool provides different configurations with varying backbone models like ResNet and PVT, demonstrating competitive inference and training times. It is built upon established frameworks such as Detectron2 and DETR, ensuring a robust and scalable architecture. SparseR-CNN is suitable for researchers and developers working on object detection, offering detailed installation and usage instructions for training, evaluation, and visualization.

SpectralCluster is a Python-based open-source library that re-implements advanced spectral clustering algorithms, particularly those used in Google's speaker diarization research. It provides functionalities for speaker diarization, including refined Laplacian matrix calculations, constrained spectral clustering, and multi-stage clustering. The tool allows users to customize various parameters such as minimum and maximum clusters, Laplacian type, refinement operations, and distance metrics for K-Means. It also supports auto-tuning for optimal performance and offers fallback clusterers for smaller datasets or specific conditions. SpectralCluster is designed for researchers and developers working on speech recognition and audio analysis, offering both standard and streaming prediction capabilities.

DebateAI is a platform designed to enhance debate skills and engage users in discussions by pitting them against AI opponents. Users can choose their opponent and a specific topic, then challenge their beliefs against AI trained to argue from any perspective. The tool provides an interactive environment for exploring different viewpoints and strengthening argumentative abilities. It offers quick start options with popular debate topics like "Are cats better pets than dogs?" or "Is remote work better than office work?" and also allows for custom debate setups. DebateAI aims to be a valuable resource for anyone looking to improve their critical thinking and argumentation skills.

mc-cnn is an open-source project hosted on GitHub, offering a comprehensive set of procedures for stereo matching tasks utilizing convolutional neural networks. It enables users to compute stereo matching costs and train CNNs for this purpose. The tool also integrates a basic stereo method that incorporates cross-based cost aggregation, semiglobal matching, left-right consistency checks, median filtering, and bilateral filtering. It requires a powerful NVIDIA GPU with at least 6 GB of memory for KITTI datasets and 12 GB for Middlebury datasets, supporting GTX Titan, K80, and GTX Titan X. The repository provides detailed instructions for installation, running pre-trained networks, training on KITTI and Middlebury datasets, and experimenting with different network architectures.

The NSF AI Institute for Societal Decision Making (NSF AI-SDM) is a research institute dedicated to advancing human-centric AI for the benefit of society. It fosters collaboration between AI and social sciences researchers to deepen the understanding of human decision-making processes. The institute applies cutting-edge AI advancements to enhance decision-making in critical areas such as resource allocation, public health, and other societal challenges. By integrating diverse expertise, NSF AI-SDM aims to create AI solutions that are not only technologically sophisticated but also ethically sound and socially beneficial, contributing to a more informed and equitable society.