Computer Vision News - February 2019

You have a new platform for annotation. What’s special about it? At Clay Sciences, we built a platform for annotating data. We specialize in video annotations, annotating directly on the video and not on single frames. First of all, one of the things that we utilize is the continuous aspect of video. Annotation on videos does not require annotating each and every frame. You can, instead, annotate single keyframes and interpolate between them. This is something that happens on our platform and saves a lot of time annotating long videos. One other thing we added to our platform is automatic tracking. We added a tool that tracks objects automatically using image processing algorithms like optical flow. This saves you even more time than annotating only keyframes. We have one video with 10 seconds - that is, 300 frames - of cars driving on Brooklyn bridge, where one annotation on the first frame gets updated automatically to the rest of the 300 frames, so that's all the work the annotators have to do. We developed tools that will annotate bounding boxes, lines, nonrigid, and rigid polygons. We have tools to annotate all of them. Our tools are very easy to use and very easy to configure. We have an interface that enables the manager to configure classes and attributes in a text box. We don’t annotate only labels. We also annotate attributes that change over time, like whether the car is braking, what’s the color of the person’s purse, their names, things like that. Both classes and attributes configured very easily through JSON format. We also manage the process of distributing the work with multiple workers. You just need to register all of the users, and everybody gets access to all of the annotation tools. Managers also get reviewing tools including contribution reports. These features are particularly precious in the autonomous vehicle industry. Can you tell us more about it? Sure, right now all of the autonomous vehicle companies record the road using videos, but still, they need to convert the videos to images because annotating videos is harder and tend to get messy. With the right tools annotating videos can get even easier. We want to move forward. We want to be more progressive instead of annotating single frames. As I said before, Guy Kohen is the cofounder of Clay Sciences, a company that develops web-based tools for annotating videos, images and text. 32 Application Computer Vision News Application “We added a tool that tracks objects automatically using image processing algorithms like optical flow”