Research Tools: code, datasets, & models
Discover an index of datasets, SDKs, APIs and open-source tools developed by Microsoft researchers and shared with the global academic community below. These experimental technologies—available through Azure AI Foundry Labs (opens in new tab)—offer a glimpse into the future of AI innovation.
Open datasheets framework
This framework aims to assist in the documentation of datasets to promote transparency and help dataset creators and consumers make informed decisions. You can read more about it in our paper:Â Open Datasheets: Machine-readable Documentation for…
BankNote-net: Open Dataset for Assistive Currency Recognition
Assistive software applications have been developed for a variety of day-to-day tasks, including currency recognition. BankNote-Net is an open dataset for assistive currency recognition.
Causal impact long-COVID
Code supplementing AI for Good Lab’s work on long COVID sequelae using EHR data. This project aimed to discover long COVID sequelae (symptoms) from Electronic Health Records (EHR) data using causal impact analysis of time series…
Identifying Whales in High-resolution Satellite Imagery
This repository contains code for an active learning pipeline for detecting whales in high-resolution satellite imagery.
Turkana grid mapping
Mapping electrical infrastructure can support measuring access to electricity and can help identify opportunities to improve or extend existing power infrastructure.
Turkana camp roof mapping
This project automates the identification of buildings and solar panels in aerial imagery, aiding humanitarian mapping. In many developing regions, maps are often outdated or missing, hindering development planning and disaster response. Our work accelerates…
Giraffe keypoint detection
This repository contains code for running a giraffe keypoint detection model. The model estimates 4 keypoint locations in images containing a single giraffe, namely: top of ossicone, top of head, neck indent, and front bottom…
nnScaler
The nnScaler is a system that takes a DNN model that designed for single device, e.g., GPU, and automatically convert it into the program that can execute concurrently on multiple devices.