Not Your Average Internship: How my Summer Internship Prepared me for Full-time Impact
After I completed my doctorate last year in Information Sciences and Technology from Penn State University, I was fortunate to be offered a full-time job working as part of the Adobe Research team. As a complement to my formal education, I had completed three internships with Adobe which provided me with a good technical foundation for work related to deep learning technology and computer vision.
Adobe’s internship program is designed to empower students and encourage them to conceptualize and develop innovative new ideas.
Besides helping me land a full-time job at the company, my internships helped me become more creative, analytical, and inquisitive. I had the opportunity to learn from very experienced researchers, and to develop new ideas with the potential for publication and product development. In fact, in collaboration with Adobe researchers, we published five papers and received or applied for seven patents. I was also fortunate to have Zhe Lin, a senior research scientist in the Adobe Research and five other researchers as my mentors.
After joining Adobe last year as a full-time research scientist, I began thinking about how deep learning technology could be used in mobile applications. The projects that I am working on now — image tagging on mobile, image auto-segmentation and image hole filling — use deep machine learning technology, which relies on algorithms to model and process large amounts of data.
One of the new technologies being developed here at Adobe Research is an auto-tagging protocol to power photo search.
The work I’m doing now builds on Adobe’s auto-tagging technology, called Haystack, which was developed by our Research Team. Auto-tagging is already being rolled-out for Lightroom on the Web, which runs on the cloud and helps users index their personal photos in an album, making it easier to retrieve a particular image. I adapted this algorithm in the mobile environment, where we have limited storage and computing resources compared with the cloud environment, and designed a mobile workflow that we are now testing.
What we are doing is developing a new algorithm to assign tags that describe the core aspects of a given image. Our goal is to automatically tag uploaded images, and provide users with a list of related images that may be of interest to them.
In the mobile environment, this tagging system automatically generates search terms based on the input image, which facilitates suggestions for other potentially relevant Adobe Stock images. The key advantage of a local tagging solution is privacy and quicker response to user’s feedback. We are currently open to both the cloud-based solution and the local solution.
I believe this technology is a perfect fit for the mobile. Soon, we will be able to offer users images from the Adobe Stock library based on the characteristics of the photo or illustration they are using at any given time.
Ultimately, this auto-tagging feature has the potential to help users find the right image for the right application, no matter where they are in the Adobe Cloud ecosystem.
Some of our most amazing tech advancements likely started with an intern or university collaboration project. Check out some of the other stories in this series, which share how interns are the secret to a thriving research lab.
You may also like: