Tight Masks for KAIST thermal pedestrian images

These are tight pedestrian masks for the thermal images present in the KAIST Multispectral pedestrian dataset, available at https://soonminhwang.github.io/rgbt-ped-detection/

Both the thermal images themselves as well as the original annotations are a part of the parent dataset. Using the annotation files provided by the authors, we develop the binary segmentation masks for the pedestrians, using the Segment Anything Model from Meta.

All masks are present as grayscale binary png files, having pixel values 255 (for the relevant masks) and 0 (everything else).

All images follow the same folder structure as the original images posted by the authors of the KAIST Multispectral Pedestrian Dataset, for ease of use.

The current dataset is a part of the work titled "An Image Information Fusion based Simple Diffusion Network leveraging the Segment Anything Model for Guided Attention on Thermal Images producing Colorized Pedestrian Masks". If you find this dataset useful, we would be grateful if you could include the citation of the research work.

We also include the data cards for the train-validation-test for the mask images used in our research work. This is just for reference for reproduction of the results. Please feel free to use any combination of the images for your work.

Code related to the dataset: https://github.com/Suranjan-G/pedestrian-masks