"Container_Dataset"

"In this study, we constructed a container defect dataset derived from real-world operational scenarios. The raw data were collected from Yantai Port, comprising 2,347 high-definition RGB images with a resolution of 1920 *1080 pixels. To ensure data diversity and generalization capability, the image acquisition process covered various lighting conditions, shooting angles, and complex background clutters at the port site. During the data annotation phase, the open-source graphical image annotation tool LabelImg was employed to perform precise bounding box annotations on the defect targets. The entire dataset contains 4,692 defect instances, covering six typical defect types: surface deformation, hole, rusty, beam deformation, roof, and scratch. The data distribution exhibits a significant long-tail characteristic. Specifically, surface deformation accounts for the largest number of samples (2,616), dominating the dataset, whereas scratch (138 instances) and roof (126 instances) are relatively scarce. This class imbalance accurately reflects the natural occurrence patterns of container defects in port environments."