SUTD Professor Sai-Kit Yeung Oversees Exciting Research During Year-Long Visit at MIT
By Jesse DeLaughter, SUTD Collaboration Office
August 8, 2014
SUTD Professor Sai-Kit Yeung has spent the last year at MIT as part of the MIT-SUTD Faculty Development Program, also known as Teach the Teachers. Through the program, Professor Yeung is experiencing the academic culture of MIT and conducting research under the mentorship of Professor Wojciech Matusik in the Computer Science and Artificial Intelligence Laboratory (CSAIL). He supervises a team of researchers whose work could eventually change many aspects of everyday life. From 3-D printed clothing to super high-resolution photographs and computers that can suggest interior design solutions, this group's projects build on their individual strengths in creative ways. Here is a glimpse into their work.
3-D Printed Clothing
3-D printers seem to be everywhere lately, and can now even be purchased by regular consumers at some home improvement stores. They can be used for creating parts by DIY hobbyists, rapid prototyping for entrepreneurs and small-scale production for manufacturers. Now, add making clothing to that list. Noah Duncan, a grad student visiting from UCLA and Junyan Wang, a visiting postdoc from SUTD, aim to solve a key problem in the production of 3-D printed clothing. Most 3-D printers have a limited volume capacity, making it impossible to print an object that extends beyond certain dimensions. Designs can be folded so as to fit the confines of the printer, but this requires the use of support materials between the folds to stabilize the structure, which adds to the cost. Enter Duncan and Wang, who use computer programming to optimize the geometry of folding to minimize the support material needed. The shirts they experiment with are made from a composite of rubber and plastic- not the most comfortable combination. But future advances could make 3-D printing a viable option for garment production, and their optimization efforts will help keep costs low.
Fashion and Design Advice from a Computer
Imagine waking up and getting fashion advice for the day with the click of a button. Craig Yu, a visiting scientist from SUTD, has developed a program that allows a user to enter certain parameters for style and color, then let the computer do the rest. It generates various occasion-appropriate outfits from a selected wardrobe. He has also used similar algorithms to arrange sets of furniture and objects within virtual spaces, creating interior designs that rival those created by human designers. In addition to potential real-world applications, the technology could be used to efficiently generate clothing on characters and furniture arrangements in spaces in computer games and virtual worlds. Both the interior design and fashion programs have been featured at the prominent SIGGRAPH conference, in 2011 and 2012, respectively. Most recently, Yu helped organize a Vision Meets Cognition workshop at the Conference on Computer Vision and Pattern Recognition in Columbus, Ohio. The workshop brought together multiple experts in the field to share new research that allows computers to better discern aspects of images formerly only interpretable by humans, including functionality (What can you do with this object?), physics (How likely is it that a balancing stone will fall), intentionality (Why is this person kicking in a door?), and causality (Who knocked down the dominos). The ability to answer such questions could lead to the creation of truly intelligent machines, with wide-ranging applications. Dr. Yu will begin as Assistant Professor at UMass Boston this September.
Super High Resolution Photography
If you have ever been awed by an enormous, larger-than-life photograph that maintains its clarity, even up close, then you are familiar with the magnificent power of super high-resolution photography. The technology is already quite good, thanks to gradual advances, which have primarily been achieved by recombining multiple images. In traditional methods, the camera's sensor, comprising rectangular pixels on a rectangular grid, is shifted, often by a fraction of a pixel, to capture additional detail. SUTD-MIT Postdoctoral Fellow Boxin Shi is working in collaboration with Professor Yeung, MIT Professor Ramesh Raskar in the Media Lab, and the Lincoln Lab, pursuing a new line of research, which focuses on varying the actual shape of the pixels.
Click here to read the paper.
Segmentation and Removal of Images from Video
In addition to supporting the work of others in the group as a technical consultant on mathematics and computer programming, Junyan Wang is involved in his own project to improve the way objects can be removed from film across multiple frames. The basic process, which descended from the rotoscoping technique, invented around 1915, involves manually tracing the contours of the object to be clipped. In the modern method, there is a three-step process. After the contours are traced, a computer program, such as Adobe, removes the object and propagates the background to subsequent frames. In the final step, the propagated frames are refined to correct any errors. Until now, Wang says, there has been very little focus on this final step. By developing better refinement techniques, performance of image segmentation programs can be improved by two to three times.
Click here for more information about CSAIL.
Click here for more information about the MIT Media Lab.
Click here for more information on the MIT-SUTD Faculty Development Program.