IAM group has won a computer vision crowdsourcing challenge

April 11, 2016 at 3:15 pm by

The Interactive and Augmented Modelling Group (IAM) of the Computer Vision Center has won a crowdsourcing challenge organized by Pallas Ludens. It is a young startup that tightly integrates crowdsourcing with computer vision.

They selected the three most interesting proposals by end of January and did a project worth 5000€ at NO COST with the winners. The money can be used either for the actual crowdsourcing tasks, for UI development carried out by our team of engineers, for quality assurance support or for consulting by the company’s CTO Daniel Kondermann. They also will publish the results together with the winners in an appropriate conference or journal.

The project that IAM presented is in collaboration with the Respiratory Endoscopy Unit at Bellvitge University Hospital. Their main research topic is the development of Intelligent In-vivo Endoscopy Navigation, Diagnosis and Intervention Support systems from the analysis of endoscopic videos using advanced image processing methods.

 What is the benefit from crowdsourcing this project?

Aside training and validation, there are three major bottlenecks in the development of new clinical decision support systems:

1. Iterating the system using as much corrections of the system output as possible made by as many clinical experts as possible.

2. Assessing the system clinical benefits in terms of improved clinical decisions and outcomes and reduction of clinicians’ learning curve once the system has been trained, validated and iterated.

3. Disseminating and promoting its use among the medical community. The use of a flexible interface allowing annotation of visual data and statistics on users, annotations/decisions times would allow to alleviate the 2 first points by using the medical communities and societies as well as non-expert general public as a crowd. Besides, a side benefit of using a crowdsourcing targeted to clinicians is to let them know about the benefits of using new processing clinical algorithms.