Experimental Comparison on Cityscapes Dataset for Thangka Image Segmentation

73.28% mIoU and 40.01% mB-Fscore were achieved by an improved PIDNet model on a custom Thangka dataset, while maintaining 109.03 FPS. The model was also evaluated on the Cityscapes benchmark to confirm generalization capability. The dataset, shared by Jiao Wu on figshare under CC-BY-4.0, was last updated in May 2026.

Use Cases

• Benchmarking semantic segmentation models based on performance metrics like mIoU and mB-Fscore.
• Training models for cultural heritage image analysis based on the described custom Thangka dataset.
• Evaluating real-time inference performance based on the reported 109.03 FPS metric.
• Testing model generalization capability based on cross-evaluation with the Cityscapes benchmark.

Strengths

• Provides concrete performance metrics: 73.28% mIoU and 40.01% mB-Fscore on a specific task.
• Includes real-time performance benchmark of 109.03 Frames Per Second.
• Dataset is openly licensed under CC-BY-4.0.

Limitations

• The dataset is very small at 5.5 KB, suggesting limited scope or content.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

figshare

Collection Method

Likely contains experimental results and performance metrics from model evaluation.

Freshness

Last updated 2026-05-14 17:41:16; freshness should be verified.