SemanticKITTI Test Set Evaluation Results for 31 Methods

Evaluation results compare the performance of 31 distinct methods on the SemanticKITTI test set. The dataset contains results for point-based, voxel-based, image-based, and projection-based semantic segmentation approaches. Xuewen Feng compiled these results, which were last updated in April 2026.

Use Cases

• Benchmarking mIoU scores across 31 different method types (point-based, voxel-based, etc.) on the SemanticKITTI test set.
• Analyzing performance trends between method categories like projection-based and point-based using reported evaluation metrics.
• Comparing the results of specific method rows (e.g., Row 5 vs. Row 25) to inform model selection for autonomous driving perception tasks.

Strengths

• Direct comparison of 31 distinct semantic segmentation methods.
• Results are derived from the standard SemanticKITTI test set (sequences 11 to 21).
• Clear categorization into four major method types for structured analysis.

Limitations

• Dataset is extremely small at 21.5 KB, containing only aggregated results, not raw data.
• The specific evaluation metrics (e.g., mIoU, accuracy) and per-class scores are not detailed in the provided description.
• Rows and columns are undefined, limiting detailed reproducibility of the analysis.

Provenance

Source

SemanticKITTI dataset benchmark.

Collection Method

Compilation of published evaluation results from various research papers or implementations.

Freshness

Last updated in April 2026.