Computer Vision News - July 2020

47 Trevor Darrell “Maybe this is a month or two to just reflect on and attend to all these very important issues.” Let’s be radical! Well, I think we have too many papers being published right now and toomany papers that are similar. When you boil down this work to its essence, there are fewer ideas, but more papers. Don’t get me wrong, they’re generally well written and have more experimental rigor now than they did in the past, but I don’t think the trend is moving in the right direction. One crazy suggestion I have is to remove the absolute requirement for novelty! There are thousands of publications in computer vision and machine learning every year and it can feel as if they all have to bring something new to get past the reviewers. Nothing is completely new in our endeavors. We might make some headway on the enormous growth of publications in our field if people got more credit for replicating existing ideas but with an important twist. The whole reward system that the community has now dissuades people from doing that. I hate to say it, but I feel like it encourages authors to make their work more complicated or obtuse than it needs to be. I was taught in elementary school that the best messages are the simplest ones, and before I became a computer vision researcher, I was taught that the best theories are too. Our publication process encourages complexity for the sake of complexity because of the review process. I’d like to encourage our field to consider how to push things in the direction of bringing simplicity and clarity to the literature. Obviously, we need new ideas, and I don’t want to throw novelty out completely, but I believe the pendulum has swung to a degree that may not be healthy. Best of CVPR 2020