The interaction on remote sensing images in natural language is a new task for the Earth observation community. Prof Devis Tuia’s team share openly a training dataset to explore this domain.
We felt that other people would like to join us and develop new ways to tackle this new problem.
What?
A training dataset designed to for remote sensing visual question answering, hosted on its dedicated website with accompanying scripts, unlicensed.
Why Open?
RSVQA is a new task for the Earth observation community. Prof. Devis Tuia and his team wanted to explore something new (the interaction on remote sensing images in natural language) and they felt that other people would like to join them and develop new ways to tackle this new problem. They also spent months designing a dataset to get started, which they think could be a barrier for many people to get started and focus here rather than on existing super-already-studied other datasets. As such, making the dataset openly available seemed a natural way to enable new progresses.
Who benefits?
It is a dataset for anyone in the remote sensing community, so scientists working with satellites and developing algorithms to extract information from them, but also, more at large, the whole computer science and computer vision communities. The original VQA task was first proposed in computer vision to help visually impaired people (who could ask questions in language regarding “things” that appear in images acquired from smarthpones and the like), but is now being extended to many other domains, such as, with Prof. Tuia’s team contribution, remote sensing.
They don’t really have statistics about usage yet, but they receive regularly emails with questions about the dataset and they see the first citations of Their paper from other studies improving on our algorithms, so it is being used!
How?
The data is hosted on the private server of Dr. Lobry. To speedup download for sometime they mirrored it on Google Drive, but they stopped. But if the data grows a lot (for now it is “only” 14.5Gb) they will have to think to something more professional.
Contact:
