Comparison Results: SSH-YOLO vs. YOLOv8n Object Detection Performance

Tenglong Ma published a dataset of comparison results for the SSH-YOLO object detection model on 2026-04-15. The data, stored in an XLS file, contains performance metrics from experiments on a self-built RoadScene-Complex dataset and four public datasets (BDD100K, KITTI, COCO, CityPersons). It is a small dataset of 5.5 KB.

Use Cases

• Benchmarking object detection models for small targets (< 32x32 pixels) based on the described four-level detection system.
• Evaluating model performance on occluded objects based on the spatial and channel collaborative attention module (SCSA) results.
• Comparing inference speed and accuracy trade-offs for real-time on-board detection based on the reported 60 FPS and mAP scores.

Strengths

• Performance metrics are provided for five distinct datasets, including a self-built RoadScene-Complex dataset.
• Specific [email protected] improvement percentages are reported, such as a 12.4% increase over YOLOv8n on the RoadScene-Complex dataset.
• The model's inference speed of 60 FPS is documented, indicating suitability for real-time applications.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset's small size (5.5 KB) suggests it contains summary results, not the underlying training or validation data.

Provenance

Source

Tenglong Ma via figshare

Collection Method

Likely contains experimental results from training and evaluating the SSH-YOLO model.

Freshness

Last updated 2026-04-15 18:05:38; freshness should be verified.