Training images for semantic segmentation for general road surface

The ZIP file "training_data_road_images.zip" includes 3200 road surface images for training. Those images include original images and their three types of symmetric images. An example of those images is as follows. c_GX0041_000877: an original image c_GX0041_000877_lr: a symmetric image about left and right of the original one c_GX0041_000877_ud: a symmetric image about top and bottom of the original one c_GX0041_000877_udlr: a point symmetric image of the original one The ZIP file "training_data_ground_truth.zip" includes 3200 ground truth images (supervised images) corresponding to their road images. Those images are gray scale (8bit) and in our data, an integer from 1 to 4 is given to each pixles. Meanings of the integers are as follows. 1: road_inside 2: road_outsides 3: manhole 4: joint etc