Adobe Snatches Up Stars from Crumbling Mitsubishi Lab—Creates Boston Research Outpost
Adobe Systems, the San Jose, CA-based company whose graphics and visual design programs are used by millions of people every day, has hired at least three prominent Boston-area computer scientists away from Cambridge’s troubled Mitsubushi Electric Research Laboratory (MERL) to form Adobe’s first significant research outpost outside the West Coast. And more MERL researchers could soon be joining their colleagues.
The group will be part of Adobe’s Advanced Technology Labs and will be physically located at the company’s existing software development facility in Newton, MA, just across the Charles River and only a few miles from Cambridge. But in spirit it will reconstitute at least part of the once-highly regarded basic research team at MERL. As Xconomy reported in July, that group has been shedding top researchers ever since Mitsubishi executives fired its director last October in a dispute over the role of basic research in the corporation and the relevance of the lab’s research to Mitsubishi’s business lines.
“We are certainly hoping it’s going to turn out to be like ‘MERL 2,’” says Paris Smaragdis, a MERL researcher snatched up by Adobe just this week. “Most of the senior people who are going to be there are actually MERL people. I think [Adobe] will be a place where we can strike a good balance between doing basic research and stuff that’s applied to products. MERL always had that balance—but [top management in Japan] just didn’t get it.”
So far, Adobe executives are treating the MERL acquisitions without fanfare. “We have hired a few people from MERL…[and] started a small outpost in Newton,” remarked senior vice president and chief software architect Tom Malloy, who heads Adobe’s Advanced Technology Labs, in a recent conversation with Xconomy. Malloy said employing researchers in the Boston area will help cement ties with local research institutions. “We’re interested in sort of pairing up with our colleagues in our Newton development office and also being partners with some of the local universities out there,” Malloy said.
The first MERL researcher to join Adobe was Wojciech Matusik, an expert in computational photography, or the use of computer algorithms to create images that are beyond the capabilities of traditional cameras. In one project with MIT computer scientist Frédo Durand, Matusik helped to develop a multi-aperture camera that can take a single photograph at several different aperture settings or f-stops simultaneously, making it possible to “refocus” a photograph after it’s been taken. Matusik, who was both a graduate student and a postdoctoral associate at MIT, accepted a position at Adobe in May.
In early August, Xconomy learned off-the-record that Matusik was being joined at Adobe by a second MERL researcher, Israeli computer vision researcher Shai Avidan. Avidan joined MERL in 2004 after several years developing image-tracking software for MobilEye, a Dutch company that builds vehicle driver-assistance systems. One of his MERL projects, a “seam carving” system for resizing and manipulating digital photographs, has received enormous attention in the last week with the posting by the technology blog TechCrunch of a striking YouTube video demonstrating the technique (original YouTube video here). Avidan’s move to Adobe has now been confirmed publicly by an Adobe employee, John Nack, who posted the information on his blog yesterday. TechCrunch picked up that news today.
Smaragdis, the latest MERL-to-Adobe transplant, studies “computational audition,” or computer listening, the use of audio cues from multiple microphones to reconstruct features or events in a 3-D space. He has used his algorithms to show that sound can be used to diagnose trouble in a machine such as an elevator, keep a motorized camera pointed at a roving subject, scour traffic-surveillance videos for the sounds of collisions or near-accidents, or separate distinct sound sources in a noisy room.
What will an audio researcher do at a company focused on images? “Adobe is doing a lot of work on video, and a lot of video comes with an audio track, and I think that’s where they see me fitting in,” says Smaragdis.
Matusik, Avidan, and Smaragdis will soon be joined at Adobe by another Paris: Sylvain Paris, a post-doctoral associate at the MIT Computer Science and Artificial Intelligence Laboratory who has developed methods for extracting 3-D information such as the geometry and detail of hair from multiple 2-D images. One of Paris’s projects is a “tone management” system for digital photographs that can, for example, give an amateur black-and-white photo the look of an Ansel Adams masterpiece. “We can analyze what makes a picture by a master look good, and take those properties and apply them to your picture,” explains Paris, who says he is already engaged in a joint research project with Matusik.
The Newton researchers will work under the direction of Adobe senior principal scientist David Salesin, a computer graphics expert who is an alumnus of Lucasfilm, Pixar, and Microsoft Research and who holds a professorship in the Department of Computer Science and Engineering at the University of Washington.
Adobe declined Xconomy’s request for an interview with Salesin, so we haven’t yet gauged how closely his management style and research philosophy resemble those of Joseph Marks, the former MERL research director credited with the MERL team’s rise to prominence. As Xconomy documented in its initial report, MERL CEO Richard Waters and managers in Mitsubishi Electric’s home offices in Japan saw a disconnect between Marks’s research agenda for the team—which was heavy on subjects such as computer vision and machine learning—and Mitsibushi Electric’s current products. Marks’s firing over the issue last October set off the unraveling of that institution’s research group.
Smaragdis says he and the other former MERL researchers are looking forward to working for a company where basic research that doesn’t necessarily focus on incremental improvements to existing products is valued. “I think that at the upper levels of Adobe they definitely have a product plan for the next five years,” he says. “But they’d also like to know what is going to happen after that, and that’s where we will sit in the company—doing the way-out-there stuff that will eventually be useful and productized, but maybe not immediately so.”
Smaragdis says Adobe has assured the MERL researchers that there will be funds to continue one of MERL’s strongest traditions: hosting a steady stream of student interns and engaging in research collaborations with local university labs. “We will also provide a good interface with the academic community, so if there is any cool stuff going on at MIT, for example, we will work with them as much as we can and get good things going.”
Adobe employees are evidently energized by the new hires. Nack, a senior product manager for Adobe Photoshop, called Matusik, Avidan, and Sylvain Paris “heavy-hitters” and “imaging science rock stars” in his blog, and said “it’s just exciting that so many smart folks are joining the team.” (At the same time, Nack emphasized that consumers won’t necessarily see technologies such as Avidan’s seam-carving algorithm included in Adobe software right away, as TechCrunch’s Michael Arrington jokingly demanded in his previously mentioned post, dubbed “I Want This in Photoshop Immediately.” As Nack wrote: “Just because a particular researcher has worked on a particular technology in his or her past life, it’s not possible to conclude that a specific feature will show up in a particular Adobe product.”)
In communications with Xconomy, Adobe has downplayed its recent recruiting binge and disavowed any plans for a more comprehensive or strategic research effort in Newton. “It would be incorrect to refer to our modest hiring in the Boston area” as the creation of a major new East Coast research center, says Eileen Foley, senior manager for corporate public relations.
“We are always looking for top talent for our Advanced Technology Labs and we recently welcomed a number of new hires—including some individuals who formerly worked for MERL,” Malloy added in response to a request for further comment. “We’re very fortunate to be able to house some of our team members in our existing office in Newton and it’s our hope that they will help us to build stronger ties with the local academic community, including MIT. They join a team that is already distributed throughout Adobe offices in San Jose, San Francisco, and Seattle.”
Adobe’s MERL hiring spree may not be over. “I suspect other MERL people might be following suit but I haven’t heard anything,” Smaragdis says. Given that the core Mitsubishi research team included only 20 scientists, just a few more hires might qualify Adobe’s Newton facility as MERL reloaded.