CULane Lane Detection Dataset

This is a large-scale dataset for traffic lane detection. The data is acquired through cameras installed on six different cars, all driven by different drivers on the streets of Beijing.

The dataset contains more than 55 hours of video, from which 133,235 frames are extracted. Among them, 88,880 images are used as training sets, 9,675 images are used as validation sets, and 34,680 images are used as test sets. The test sets are divided into common classes and 8 challenge classes.