Mural Damage Recognition and 3D Semantic Mapping Dataset with 6,266 Images

6,266 annotated mural images form a multi-source dataset for detection, segmentation, and 3D validation of damage types. The dataset was constructed by Hu Yanfeng and last updated on 2026-05-02. It supports a proposed framework for 2D damage recognition and 3D semantic mapping of mural heritage.

Use Cases

• Training deep learning models for multi-class mural damage detection based on the described YOLO-MME framework
• Performing fine-grained segmentation of cracks, flaking, and crazing based on the improved DeepLabv3+ network mentioned in the description
• Creating 3D semantic maps of mural surfaces for spatial damage visualization and measurement
• Validating 3D reconstruction and label fusion pipelines for heritage monitoring applications

Strengths

• Contains 6,266 mural images specifically constructed for detection, segmentation, and 3D validation tasks
• Framework performance metrics are provided, with the YOLO-MME model achieving an F1 score of 88.4% and mAP@50 of 90.8%
• Dataset is 1.4 GB in size, indicating a medium-scale collection of image data

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment for specific modeling tasks
• Geographic and institutional provenance of the mural images is not specified, which may affect generalizability

Provenance

Source

Hu Yanfeng via figshare

Collection Method

Constructed as a multi-source dataset for a proposed deep learning framework; likely involves image collection and annotation.

Freshness

Last updated 2026-05-02 04:35:06; freshness should be verified