Data

The images are taken from roughly the same spot, about 55 meters above ground, over more than 5 years. Most pictures face roughly the same angle south-east (Boston Skyline), and a few face south-west (Cambridge - MIT Campus). Several cameras were used: Canon 5D Mark II (21 MP), a GoPro Hero 2, a GoPro Hero 3, and several point-and-shoot and phone cameras. Most pictures are JPG format, sometimes bracketed for extended dynamic range, with a few RAW images and videos. Some pictures are singletons (just one picture taken in a few days), but many are part of timelapses (pictures taken every second or minute). Statistics about picture acquisition is available on the right and in the graph below.

Statistics

Here are some relevant statistics about the dataset. Please contact us if there's anything else you'd like to see in this table.

Property	Approx. count
Boston skyline pictures	1.028 million
Cambridge/MIT campus pictures	0.205 million
JPGs	1.33 million
RAWs	14 thousand
Videos	100
No. months covered.	71

Data Visualization

tipiX is a tool that facilitates the quick and interactive exploration of image datasets. We've prepared two tipiX frames with tiny sample of the Boston dataset:

Exploration 1: a few untouched images, randomly (roughly uniformly over time periods). This sample will be roughly indicative of the quality of the data in terms of camera angle, lighting, etc - including pictures that don't belong but were included by accident. The images have been rescaled to 540 x 960px.
Exploration 2: curated and aligned sample.

Exploring the data in this manner has allowed us to find interesting events in Boston, such as the 2012 transformer fire. If you'd like to work with this data, please don't hesitate to contact Adrian Dalca.

Research

Given the size and quality of the dataset, there are many interesting directions to explore. We've found some interesting phenomena, including the peculiar daily visual effect shown below. Can one predict the temperature based on a Boston image? Can one synthesize timelapses under specific parameters? Can the Canon 5D pictures improve the (more available) GoPro Hero pictures? Can we estimate the Boston Skyline depth with the parallax and shading? If you're interested in any research problem and can use this data, simply contact me. In the future, we hope to make the dataset easily available to download, but a lot more pre-processing and organization needs to happen until then.