AutoRefuel-Seg: An Instance Segmentation Dataset for Fuel Tank Door Flap and Inner Cap Knob Perception

AutoRefuel-Seg is a real-world instance segmentation dataset constructed for robotic automotive fuel tank component perception. It contains RGB images of two key fuel tank components: the fuel tank door flap and the fuel inner cap knob. The images cover different vehicle types, body colors, observation scales, and lighting conditions, including daytime and illuminated nighttime scenes. The dataset provides polygon mask annotations in YOLO instance segmentation format, where each annotation includes the class index and normalized polygon coordinates. The dataset contains 3,240 images with corresponding polygon mask labels. Before release, the longer side of each image was resized to 640 pixels, while the shorter side was scaled proportionally. This dataset is associated with the manuscript submitted to The Visual Computer and is intended to support reproducible research on lightweight instance segmentation and robotic fuel tank component localization.