Ablation Experiment: Performance Metrics for DSE-YOLO11 Traffic Detector

Experimental results from 2026 evaluate a lightweight object detector for complex road scenes. The dataset, created by Yange Chen and shared under CC-BY-4.0, compares the proposed DSE-YOLO11 model against baselines on the RAD dataset. It likely contains quantitative performance metrics such as recall and mAP50, as well as model complexity figures like parameter count and GFLOPs.

Use Cases

• Benchmarking lightweight object detectors based on the reported recall and mAP50 metrics.
• Analyzing the trade-off between model performance and complexity using the provided parameter and GFLOP counts.
• Evaluating detection robustness for small and low-contrast traffic elements in complex scenes as described.
• Comparing integration strategies for dynamic convolution, lightweight neck architectures, and attention mechanisms in object detection pipelines.

Strengths

• Includes specific performance improvements: recall increased from 0.744 to 0.811 and mAP50 from 0.810 to 0.856.
• Reports concrete model complexity metrics: 2.96M parameters and 7.1 GFLOPs.
• Provides preliminary external validation on the BDD100K dataset as mentioned in the description.

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 5.5 KB size indicates it contains summary results, not the raw image or annotation data.

Provenance

Source

Yange Chen

Collection Method

Likely contains results from ablation experiments comparing model architectures on the RAD and BDD100K datasets.

Freshness

Last updated 2026-06-01 17:34:28; freshness should be verified.