This dataset is designed for training deep learning models in urban flood depth estimation using street-level images of submerged vehicles. The dataset consists of 3,367 images categorized into five flood levels based on water height relative to vehicle features (e.g., tires, chassis, windows). The images were extracted from an original dataset (https://github.com/ywszxdd/original_dataset, reference paper: Automatic detection of urban flood level with YOLOv8 using flooded vehicle dataset), cropped using original bounding boxes, with a 20% enlargement around the detected objects. Additionally, all images were resized so that the smaller dimension is at least 224 pixels, with a maximum scaling factor of 4×, using Enhanced Deep Residual Networks for Single Image Super-Resolution models. The dataset is divided into five folders, each corresponding to a specific flood depth level. Additionally, three text files define the training, validation, and test splits. Split Level 0 Level 1 Level 2 Level 3 Level 4 Total Train 640 564 892 356 240 2,692 Validation 80 70 111 44 30 335 Test 80 72 113 45 30 340 Total 800 706 1,116 445 300 3,367