Shulou(Shulou.com)06/01 Report--

MIT and Toyota release for self-driving research data set DriveSeg is how, many novices are not very clear about this, in order to help you solve this problem, the following editor will explain for you in detail, people with this need can come to learn, I hope you can gain something.

DriveSeg contains accurate pixel-level representations of many common road objects and drives shots of the scene through continuous video.

How can we train autopilot models to deepen our understanding of the world around us? Can computers learn from past experience to identify future patterns to help them safely cope with new and unpredictable situations?

Researchers at the Massachusetts Institute of Technology's Transportation and Logistics Center AgeLab and Toyota's Cooperative Safety Research Center (CSRC) released an innovative open data set called DriveSeg.

By releasing DriveSeg, MIT and Toyota are working to advance research on autopilot systems, which treat the driving environment as a continuous flow of visual information, just like human perception.

To date, the autopilot data provided to research groups consist mainly of a large number of static single images that can be used to identify and track common objects found in and around roads, such as bicycles, pedestrians or traffic lights, by using a "bounding box". By contrast, DriveSeg contains more accurate pixel-level representations of many of the same common road objects, but drives scenes through continuous video. This type of panoramic segmentation is particularly useful for identifying more amorphous objects (such as road construction and vegetation) that do not always have this definition and unified shape.

According to Sherony, video-based driving scene perception provides a data stream that is more similar to dynamic, real-world driving. It also enables researchers to explore data patterns over time, which may drive advances in machine learning, scene understanding and behavior prediction.

DriveSeg is provided free of charge and can be used by researchers and academia for non-commercial purposes. The data consists of two parts (manual and semi-auto). DriveSeg (manual) is a 2: 47-second high-resolution video captured during the day on the busy streets of Cambridge, Massachusetts. The 5000 frames of the video are densely annotated using manual labels per pixel for 12 types of road objects.

DriveSeg (semi-auto) is 20100 video frames (67 10-second video clips) extracted from the MIT Advanced vehicle Technology (AVT) Alliance data. DriveSeg (Semi-auto) and DriveSeg (manual) have the same pixel-level semantic annotations, except that annotations are done through a new semi-automatic annotation method developed by MIT. Compared with manual annotations, this method takes advantage of manual and computational efforts to annotate data more effectively and roughly at a lower cost. The dataset was created to assess the feasibility of annotating various real-world driving scenarios and to assess the potential to train vehicle perception systems on pixel tags created by AI-based marking systems.

Is it helpful for you to read the above content? If you want to know more about the relevant knowledge or read more related articles, please follow the industry information channel, thank you for your support.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.