Performance under Sparse PC Conditions: 3D Object Detection Results on KITTI and NuScenes

Nan Zhang's research dataset contains performance metrics for a novel 3D object detection method evaluated on the KITTI and NuScenes benchmarks. The dataset, last updated in June 2026, includes segmentation and detection accuracy percentages, processing times, and frames-per-second metrics. The method integrates a Cloth Simulation Filter, an improved Euclidean clustering algorithm, and an enhanced PointNet architecture.

Use Cases

• Benchmarking new 3D object detection models based on reported segmentation and detection accuracy metrics.
• Analyzing the trade-off between speed and precision in point cloud processing based on the reported processing times and frames-per-second.
• Studying the effectiveness of multi-scale grouping and multi-resolution grouping features for local feature extraction in sparse data.

Strengths

• Reported segmentation accuracies of 94.96% and 93.12% on KITTI and NuScenes benchmarks.
• Achieves real-time detection speeds of 34 fps and 31 fps on the respective datasets.
• Performance improvements are reported as statistically significant (p < 0.001) over a standard PointNet baseline.

Limitations

• Dataset is very small at 5.5 KB, suggesting it contains summary results rather than raw sensor data.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for certain analyses.

Provenance

Source

Nan Zhang via figshare.

Collection Method

Results from evaluating a proposed 3D object detection method on public benchmarks.

Freshness

Last updated 2026-06-01 17:50:52; freshness should be verified.