Google on Monday publicly released a massive image dataset of burst photography, having said the move is the latest step in its efforts to support the research community. The library consists of 3,640 full-resolution bursts coming in the form of RAW (DNG) files, with the photographs being composed from 28,461 individual images using the Alphabet-owned firm’s HDR+ algorithms. The Mountain View, California-based tech giant is providing the dataset under the Creative Commons license, meaning capable researchers and engineers could use it to develop commercial HDR+ solutions. The database can be downloaded by referring to the banner below and is over 750GB in size.
All of the images in the dataset have been captured using the publicly available Android Camera2 API running on a number of Android-powered devices from the Nexus and Pixel lineups. Google credits the cameras of the Nexus 5, Nexus 6, Nexus 5X, and the Nexus 6P for their roles in the process of compiling the library, having also revealed some images have been captured using the original Pixel and Pixel XL released in late 2016. Researchers are able to analyze individual shots based on their metadata which will reveal the exact camera model that captured them. The vast majority of the images contained in the dataset have been recorded using the same exposure time and gain settings, in addition to spanning a wide variety of scenarios, Google said.
The public release of the database should accelerate any independent HDR+ burst research in the future, allowing engineers to quickly compare their own solutions to those created by Google or even base their technologies on the company’s dataset. Google’s critically acclaimed imaging techniques implemented into the Pixel and Pixel 2 smartphone lineups largely rely on HDR+ burst to deliver photos, compiling every single shot out of a collage of images taken in a short succession but at different exposure levels. Once stitched, the photographs produced through such an approach should boast a higher dynamic range than anything you could get from the same hardware using a more traditional shooting method, with the technique hence effectively allowing users to surpass the physical limitations of their mobile cameras, at least to a certain degree.
Google's HDR+ Burst Dataset
