Side-Scan Sonar Pipeline Detection Model Ablation Results

Ablation experiment results from 2026 by Dandan Liu, evaluating modifications to the YOLO11n-seg model for detecting submarine pipelines and cables in side-scan sonar imagery. The 5.5 KB Excel file contains performance metrics comparing the improved model against baseline versions like YOLO12n-seg and YOLO13n-seg. Results demonstrate specific improvements in recall and mAP for both object detection and instance segmentation tasks.

Use Cases

• Compare bounding box recall and mAP@50-95 metrics between the modified YOLO11n-seg and original model to evaluate detection performance gains.
• Analyze instance segmentation recall and mAP@50 improvements to assess the model's effectiveness for pixel-level target delineation.
• Review the 15% reduction in model parameter size documented in the results to inform decisions on model lightweighting for edge deployment.
• Benchmark the modified model's mAP@50 performance against YOLO12n-seg and YOLO13n-seg for object detection tasks on the Marine-PULSE dataset.

Strengths

• Performance metrics are quantified with specific percentages, such as a 9.7% improvement in bounding box recall and a 1.6% gain in mAP@50-95.
• Results are directly comparable across multiple YOLO model variants (YOLO11n-seg, YOLO12n-seg, YOLO13n-seg) for object detection.

Limitations

• The dataset is extremely small at 5.5 KB, containing only summary experimental results without the underlying image data or raw predictions.
• The temporal and geographic scope of the underlying Marine-PULSE evaluation data used to generate these results is not specified.

Provenance

Source

Dandan Liu via figshare.

Collection Method

Results from ablation and comparative experiments on the Marine-PULSE submarine pipeline and cable dataset.

Time Range

null

Freshness

Last updated in April 2026, indicating recent experimental work.

Geography

null