CAVS Semantic Segmentation Dataset for Off-Road Autonomous Driving (CaSSeD) *************************************************************************************************** Authors: Lalitha Dabbiru, Suvash Sharma, Kofi Ennin, Chris Goodin, Daniel Carruth, Matthew Doude, Christopher Hudson, John Ball Corresponding Author: lalitha@cavs.msstate.edu *************************************************************************************************** If you use this dataset for research, please consider citing our dataset paper: Dabbiru, L., Sharma, S., Ennin, K.A., Goodin, C.T., Hudson, C.R., Doude, M., Carruth, D.W., & Ball, J.E. "CAVS semantic segmentation dataset for off-road autonomous vehicles.", Proc. SPIE 13474, Autonomous Systems: Sensors, Processing, and Security for Ground, Air, Sea, and Space Vehicles and Infrastructure (14 April 2025). *************************************************************************************************** The package named "CaSSeD.tar.gz" consists of real world and synthetic datasets. It consists of raw images and the corresponding annotations along with rosbags. It can be extracted in Linux with the command: tar -xvf CaSSeD.tar.gz In Windows, it can be extracted simply with any unzipping software. The "Real_World_data" folder contains five rosbags from which the training and testing images were extracted. The offroad dataset contains 1077 real-world RGB camera images and corresponding annotations. The "Synthetic_Data" folder contains 1947 images and annotations generated using the Mississippi State University Autonomous Vehicle Simulator (MAVS). Real-world Dataset Description: ------------------------------- The real-world dataset contains data collected on different trails with different environment and lighting conditions. All these off-road datasets were collected on the CAVS proving ground. The trails and the number of images with annotations were given in the table below. The "Brown_field" (200 images) and "Main_Trail" (200 images) folders contain images from trails through dense vegetation. The "Powerline" (99 images) images are from a trail running underneath a power transmission line and includes the powerline cables and poles, a dirt trail, shrubs, and trees. The "Fogdata" folder (279 images) was collected on the "Main_Trail" in foggy conditions and the "NorthFarm" folder (299 images) was collected on gravel roads running between agricultural fields. The "Train" folder contains 350 real world images and the corresponding annotations. The "annotations" folder contains another folder called "int_maps" where the integer mapped pixel values are assigned for each class. The "Test" folder contains 150 real world images taken from three different trails, 60 images from "Brown_field", 60 images from "Main_trail", and 30 images from "Powerline" trail and are saved in the "mixed" folder. "int_maps" are also provided for each test set. Trail Name # of Images with annotations Brown_field 200 Main_Trail 200 Powerline 99 Fogdata 279 NorthFarm 299 Train 350 Test 150 Annotation Description: ----------------------- The data are semantically labeled using 6 classes. They are: 1. Smooth trail, 2. Rough trail, 3. Small vegetation (like grass, shrubs), 4. Forest / Trees, 5. Sky, and 6. Obstacles The annotations of these images have unique RGB color for each class and these are represented as below: Class Color and RGB Value Int_ID Smooth Trail Gray - [155, 155, 154] 0 Rough Trail Brown - [138, 87, 42] 1 Small Vegetation Light Green - [208, 254, 157] 2 Forest Dark Green - [59, 93, 4] 3 Sky Blue - [73, 143, 225] 4 Obstacle Dark Pink - [183, 21, 123] 5 Synthetic Dataset Description ----------------------------- The "Synthetic_Data" folder contains a synthetic dataset for semantic segmentation in an unstructured driving environment generated using an in-house, physics-based, high-fidelity simulator, the Mississippi State University Autonomous Vehicle Simulator (MAVS). This set consists of 1,947 image pairs featuring different time of the day, lighting condition, plant density, cloud cover, and landscape.