We are really excited to announce that we will be organizing a workshop at CVPR 2019 this summer! We plan on giving research awards and travel grants for people from developing regions to attend CVPR. If you want to stay informed about our calls for proposals, papers and travel grants add your email in our list.
Are you a non-profit or company interested in supporting or sponsoring our initiative? Do your computer vision projects have social impact on developing regions? Reach out to us!.
Computer vision technology has recently made rapid progress in several subareas (e.g., object segmentation, video classification), achieving a level of performance that was unexpected just a few years ago. This technology has opened possibilities in many real world domains including transportation (self-driving cars, drones), entertainment (shopping, phones), safety (security systems), which are used on a daily basis to make our lives more efficient, richer, and safer. While these applications give meaning and value to our technology, their focus on certain geographical regions or markets, often in first world countries, injects biases in our datasets, tasks, and ultimately direction of the progress of the field.
Parallel to this, penetration of internet and mobile device usage in developing countries has largely increased in the past years; more people today have access to mobile phones than to piped water supply. Access to network connected computing devices which often have a camera offers computer vision a unique opportunity to bypass the lack of infrastructure and deliver expertise, advice, or information to those in need. Examples of such applications are be in farming, health care, or clean water, among others. These applications, however, often pose challenges that are not necessarily addressed in existing computer vision practices sufficiently. Dealing with datasets with a diversity matching that of the real world, addressing problems falling in the long tail distribution of data availability, or reduction of biases towards certain classes/tasks/ethnicities are some of those challenges, just to name a few. Thus, widening the scope of computer vision to address such problems could lead to a two way advantage: vision could positively impact the lives of 6 billion people in developing regions, and that could reveal some of the blind spots and biases in our current computer vision datasets, tasks, practices, and ultimately lead to purely technical progress.
Why has this opportunity not been seized before? We argue that one of the main obstacles is the disconnection between domain experts: those who know about the problems, and those who know about technical solutions. This disconnection may be geographical, may be of language, and may even be due to lack of a forum to find each other. We propose this initiative with the purpose of bridging the gap between these two communities. In particular our goals are: