PavementCrackDetection

Crack and Sealed Crack Dataset

We developed a highway asphalt pavement dataset containing 10,400 images captured by a highway condition monitoring vehicle with 202,840 labeled crack and sealed crack instances.

Please pay attention to the disk capacity when downloading.

All images and labels contain all the 10400 images and their labels.

Val is just the validation set that produced the results of our experiments.

Trained Models

On the dataset mentioned above, we trained 13 currently prevalent object detection models from scratch, and the trained weights can be downloaded.

| Model(source) | Trained Weights(on our dataset) | | ------------------------------------------------------------ | ------------------------------------------------------------ | | fasterrcnn_resnet50_fpn | link | | fasterrcnn_resnet50_fpn_v2 | link | | fasterrcnn_mobilenet_v3_large_fpn | link | | fasterrcnn_mobilenet_v3_large_320_fpn | link | | fcos_resnet50_fpn | link | | retinanet_resnet50_fpn | link | | retinanet_resnet50_fpn_v2 | link | | ssd300_vgg16 | link | | ssdlite320_mobilenet_v3_large | link | | yolov5n | link | | yolov5s | link | | yolov5m | link | | yolov5l | link |

All trained models are saved as checkpoints and could be loaded:

import torch
import torchvision
# model
model = torchvision.models.detection.fasterrcnn_mobilenet_v3_large_320_fpn(num_classes=3, box_score_thresh=0.25, box_nms_thresh=0.5)
# load checkpoint
checkpoint = torch.load("./path/to/checkpoint.pth", map_location="cpu")
# load trained weights
model.load_state_dict(checkpoint["model"])

Citation

@article{yang2022efficient,
  title={An Efficient Method for Detecting Asphalt Pavement Cracks and Sealed Cracks Based on a Deep Data-Driven Model},
  author={Yang, Nan and Li, Yongshang and Ma, Ronggui},
  journal={Applied Sciences},
  volume={12},
  number={19},
  pages={10089},
  year={2022},
  publisher={MDPI}
}