Pau Rodríguez and Suman Ghosh are participating at this year’s Google Summer of Code within the OpenCV foundation. The Google Summer of Code is a program in which Google gives stipends to students in order to contribute to more than 200 open source.
Pau Rodríguez, Research member of the Image Sequence Evaluation team is implementing state of the art computer vision and deep learning algorithms. Suman Ghosh, research member of the Document analysis team is working on text detection in images. Both, contributing with their work to the Open Source Computer Vision Library (OpenCV).
Specifically, Pau Rodríguez will be developing an API for recurrent neural networks into tiny-dnn, being his mentor at the project, Edgar Riba, who is also a CVC PhD student and has been working at Google Summer of Code for more than three years now. As Pau states “I’m working on tiny-dnn, which is smaller than other frameworks such as Tensorflow or Pytorch, thought for smaller hardware and for direct deploy”. “RNNs are now central to many applications”, he continues “from speech recognition to Computer Vision. Some examples are Image captioning, Visual Question Answering (VQA), autonomous driving, and even Lip reading.”
On the other hand, Suman Ghosh explains that he is working “to deliver a framework in OpenCV which can be used to detect and recognise text of any image”. A crucial task in order to give computers the ability to read text under different and uncertain circumstances. As Suman explained in his proposal, “End-to-end scene text recognition is usually divided in two different sub tasks: word detection and word recognition. Currently, OpenCV text detection does not use state of the art deep network(s), therefore my proposal is to implement a deep network for text detection”.
Edgar Riba is Pau Rodríguez’s mentor within the project. He is also a CVC PhD student who is currently at Arraiy, inc. at Palo Alto, California, within a research stay. After participating at Google Summer of Code for three years, he has now been elected as a mentor. Last year, he extended tiny-dnn to OpenCV, which now is totally integrated and hosted by OpenCV. In addition, he is mentoring two more students fully dedicated to tiny-dnn at this year’s Summer of Code.
“Tiny-dnn is a library that implements deep learning and is not intended for research, but for deployment, it can be suitable for limited computational resource, embedded systems and Internet of Things (IoT) devices, and it is now supported by the OpenCV foundation,” Edgar explains. Edgar, Pau and the other students are working hard on preparing tiny-dnn so it can be deployed in Raspberry Pis, within Android devices, etc. Two more students are involved to make it work with GPUs to speed up inference during deployment and preparing a model zoo in order to import trained models independently from the framework used to train.
As the program announces itself, the Google Summer of Code is a global program focused on introducing students to open source software development. Students work on a 3 month programming project with an open source organization during their break from university. The program started in 2005 and the statistics are astonishing, with more than 13.000 students involved.