"Metallic Foreign Objects in Road Scenes"

"To evaluate the detection capability and generalization potential of the proposed model in real road environments, an in-house dataset of discarded metallic foreign objects for roads and slope-protection areas was constructed. The images in the dataset were obtained by extracting frames from videos captured using handheld devices. The video acquisition covered selected sections of National Roads G216, G219, and G317 in the Xinjiang\u2013Tibet region. The recorded content mainly focused on scenes of road surfaces and roadside slope-protection areas. Data collection was primarily conducted during the noon and afternoon periods. After frame extraction, all video data were uniformly merged and organized, and low-quality and duplicate samples were removed through manual screening. To ensure annotation consistency and standardization, all images were manually annotated using the LabelImg tool. A total of 1722 targets were labeled. The task was defined as single-class object detection, with emphasis on detection performance under variations in target scale, appearance diversity, and road-background interference.The uploaded files are part of the dataset."