ThermalTrack
- Citation Author(s):
-
Yiming Yang
(Department of Electrical and Computer Engineering, Michigan Technological University, Houghton, MI, USA)
Jeremy Bos
(Department of Electrical and Computer Engineering, Michigan Technological University, Houghton, MI, USA)
- Submitted by:
- Yiming Yang
- Last updated:
- DOI:
- 10.21227/5nq0-ye10
- Data Format:
- Links:
5 views
- Categories:
- Keywords:
Abstract
ThermalTrack is an RGB-LWIR paired dataset of wheel tracks captured under harsh winter conditions, including white-outs (severely degraded visibility), low-contrast snow terrain, and diverse wheel track geometries. Designed to enable robust alternative navigation strategies for winter autonomy systems, this dataset builds upon WADS (https://digitalcommons.mtu.edu/wads/), a specialized dataset for autonomous vehicle research in inclement winter weather. To improve model generalization, we augmented the dataset using random flips, rotations, brightness/contrast adjustments, scaling, noise injection, and perspective transformations - simulating a broader range of challenging conditions. Starting with 951 base images, we expanded the dataset to 7,608 samples, enhancing the model's ability to handle real-world variability.
Instructions:
This dataset follows a strict structure of neural network training: training, validation, and testing. The data is divided and packaged into the 3 parts and ready to use. For the 3 parts, each has two components: input images and labels. Input images are 4-channel fused RGB-thermal images with size 512 x 512 pixel. Labels are also 512 x 512 pixel images.
Dataset Files
- The validation images for neural network validation. The dataset includes 512 x 512 pixel fused 4-channel RGB-Thermal images. (Size: 129.07 MB)
- This file includes 512 × 512 pixel image labels for neural network testing. (Size: 403.46 KB)
- The file includes 512 × 512 pixel fused RGB LWIR images for neural network training - sequence 1. (Size: 959.45 MB)
- The file includes 512 × 512 pixel fused RGB LWIR images for neural network training - sequence 2. (Size: 950.7 MB)
- The file includes 512 × 512 pixel fused RGB LWIR images for neural network training - sequence 3. (Size: 948.68 MB)
- The file includes 512 × 512 pixel fused RGB LWIR images for neural network training - sequence 4. (Size: 957.85 MB)
- The file includes 512 × 512 pixel fused RGB LWIR images for neural network training - sequence 5. (Size: 954.07 MB)
- The file includes 512 × 512 pixel fused RGB LWIR images for neural network training - sequence 6. (Size: 952.81 MB)
- The file includes 512 × 512 pixel fused RGB LWIR images for neural network training - sequence 7. (Size: 949.11 MB)
- The file includes 512 × 512 pixel fused RGB LWIR images for neural network training - sequence 8. (Size: 950.84 MB)
- The file includes 512 × 512 pixel fused RGB LWIR images for neural network training - sequence 9. (Size: 949.9 MB)
- The file includes 512 × 512 pixel fused RGB LWIR images for neural network training - sequence 10. (Size: 941.74 MB)
- The file includes 512 × 512 pixel fused RGB LWIR images for neural network training - sequence 11. (Size: 366.87 MB)
- This file includes 512 × 512 pixel image labels for neural network training. (Size: 11.01 MB)
- The file includes 512 × 512 pixel fused RGB LWIR images for neural network validation. (Size: 129.36 MB)
- This file includes 512 × 512 pixel image labels for neural network validation. (Size: 407.79 KB)
