MSR Student Project Relates Well to Upcoming Internship

Suhail Pallath Sulaiman’s work on synthetic dataset generation for machine learning helped him land an internship with HERE Technologies

Suhail Pallath Sulaiman illustrates 3D Scanning using an RGBD Camera

The question seemed to surprise Suhail Pallath Sulaiman.

Sulaiman, who currently is pursuing his Master of Science in Robotics at Northwestern Engineering, was in an interview for a deep learning research intern position with HERE Technologies. The interviewer had previously reviewed Sulaiman’s winter independent project for the MSR program, which focused on synthetic dataset generation for machine learning, and asked how many people were involved on the project.

Sulaiman said he put it together by himself. By the end of the interview, the internship was his.

“When I said it was an individual project, he seemed pretty impressed,” Sulaiman said with a laugh.

HERE is an Open Location Platform company that enables people, enterprises and cities to harness the power of location. The company’s goal is simple, yet ambitious: “to create a digital representation of reality to radically improve the way everyone and everything lives, moves and interacts.”

Suhail Pallath SulaimanIn the internship, Sulaiman will be preparing data sets from the data HERE has collected and training custom neural networks they have developed. This work is directly in line with the problem he tried to solve with his project.

“One of the main problems with machine learning is preparing data sets,” Sulaiman explained. “There is a lot of manual labor involved. My project was trying to automate the process.”

Sulaiman used an RGB-D camera that produces both color and distance readings that allowed him to scan an object or person and make a 3D color model. Using that model, his software generates millions of fake images of the object from different angles and with different lighting and backgrounds. These images are processed by the software to generate a dataset that can be used for training object detection neural networks without any extra work. These trained networks will then be able to identify and localize the objects in real-time camera feeds.

“This has the potential to make the work of software engineers much easier,” said Sulaiman, who acknowledged this concept of generating synthetic datasets is not new. “This could be scaled and reduce the amount of manual labor and time needed to prepare datasets for the wide variety of objects that could require training for recognition.

“Manually, this process could easily take three or four hours. With this, you could do it in 10 minutes.”

Sulaiman credited the MSR program for providing students with opportunities to solve problems facing the robotics industry, as well as the training necessary to become leaders in the area post graduation.

“We are always thinking about robotics, nothing else,” Sulaiman said. “It can be pretty stressful, but our knowledge is increasing at an exponential rate.

“After this program, I feel I’ll be pretty prepared for the industry or for starting my own business.”

Suhail Pallath Sulaiman is a current student in the MSR program.