Computer Vision News - November 2016

Jitendra concluded his lecture at the Video Segmentation workshop by showing this hilarious video, which is intriguing because it illustrates the fundamental dilemma facing us: the low-level cues (color and movement) conflict with the high-level cues of what does the person look like. Segmenting individual people might conflict with color grouping cues and with common movement. In an earlier generation, we didn’t think of it so much as a set of engineering problems to be solved. We thought of it more as scientific challenges. I think that’s the trade- off. The current generation is too much engineering and too little science, and the old generation was too much science and too little engineering, because we didn’t have any successes in practical settings. Ralph: Did you see any of your colleagues or students who managed to have both? Jitendra: I think now everyone is trying to do both. I would expect that the field evolves. Mechanical engineering is grounded in physics, but a lot of the stuff that was once upon a time subjects that physicists like Newton and so forth studied, is really in the province of engineering. Physics professors don’t study that anymore. Fields go through this transition… sometimes over hundreds of years. In our field, maybe it’s over the next 50 years. 50 years have passed and there will be another 50 years to go. The field as a whole has to evolve from science to engineering, but it doesn’t happen simultaneously for all of the problems in the field. Some problems are more at the science stage, some problems are more at the engineering stage. What I try to do is to think about when a problem becomes very much in the engineering state, then I want to get out of that problem. I want to focus on the problems which are in the earlier stages where it’s really about scientific research, because once it becomes more of an engineering problem, at that point it really should be done by industry. Ralph: So at this point you are more fascinated by science than by engineering? Jitendra: Yes! “The current generation is too much engineering and too little science, and the old generation was too much science and too little engineering” “Once a problem becomes more of an engineering problem, at that point it really should be done by industry. ” 22 Interview with Jitendra Malik BEST OF ECCV