SynPhoRest - Synthetic Photorealistic Forest Dataset with Depth Information for Machine Learning Model Training

<strong>SynPhoRest </strong>is a synthetic dataset collected on virtual forests. It features RGB images, semantic segmentation maps, depth maps and the projection of LIDAR point clouds on the RGB FOV for two different LIDAR scanning patterns. The dataset has a total of 3154 frames.<br> <br> A description of the available data follows: <strong>RGB images</strong> Resolution: 848 x 480 pixels. PNG files with 8 bits encoding per channel. <strong>Segmentation Maps</strong> Resolution: 848 x 480 pixels. PNG files with a single 8 bit channel. Classes are encoded as follows: <strong>Value</strong> <strong>Class</strong> 0 Background 1 Soil 2 Traversable 3 Canopy 4 Fuel 5 Trunks Fuel represents flammable material such as shrubbery and grass. <strong>Depth Maps</strong> Resolution: 848 x 480 pixels. PNG files with a single 16 bit channel The depth value is encoded in the unsigned integer format. Infinite depth is represented by the value 65535. To obtain the depth values in meters, the original values must by divided by 256. The FOV of the virtual depth camera was the same as the FOV of the RGB camera. <strong>LIDAR Point Cloud Projections on the RGB Camera FOV</strong> Resolution: 848 x 480 pixels. PNG files with a single 16 bit channel. The distance values are encoded in the unsigned integer format. To obtain the distance values in meters the original values must by divided by 256. On average, the projection images have a point density of 5.5%. In practice, this means that 5.5% of the pixels in the projection image have distance information. Two LIDAR Point Cloud Projections were made available. One for a LIDAR with a repeating line pattern and other resembling the commercially available Livox Horizon LIDAR scanner.