FirESAM Semantic Segmentation Dataset

FirESAM Semantic Segmentation Dataset (also referred to as FSSSD/F3SD) is a fire semantic segmentation dataset built from the FASDD_CV subset of FASDD. Bounding-box labels from FASDD_CV were used as prompts to generate pixel-level fire masks, followed by review and re-annotation to produce the final dataset.

Folder structure:

F3SD/
all_images/
all_masks/
annotations/
images/
masks/
re_annotations/
images/
masks/

See the README.md file for more info about the zipped folder.

This dataset is tied to the manuscript "FirESAM: An Ultra-Lightweight Prompt-in-the-Loop Distillation Model for Real-Time Fire Segmentation on Edge Devices and the FSSSD Dataset."