Container damage detection using advanced computer vision model Yolov12 vs Yolov11 vs RF-DETR A comparative analysis

TL;DR

This paper compares the performance of three advanced computer vision models—Yolov12, Yolov11, and RF-DETR—in detecting container damage, using a dataset of 278 images to evaluate their accuracy and suitability for safety inspections.

Contribution

The study provides a comparative analysis of Yolov12, Yolov11, and RF-DETR for container damage detection, highlighting their strengths and weaknesses in different damage scenarios.

Findings

Yolov11 and Yolov12 achieved higher mAP scores (~82%) than RF-DETR (~78%).

RF-DETR outperformed others in detecting less common damages with higher confidence.

Yolov12 and Yolov11 are more accurate for common damage detection, while RF-DETR excels in rare damage cases.

Abstract

Containers are an integral part of the logistics industry and act as a barrier for cargo. A typical service life for a container is more than 20 years. However, overtime containers suffer various types of damage due to the mechanical as well as natural factors. A damaged container is a safety hazard for the employees handling it and a liability for the logistic company. Therefore, a timely inspection and detection of the damaged container is a key for prolonging service life as well as avoiding safety hazards. In this paper, we will compare the performance of the damage detection by three state-of-the-art advanced computer vision models Yolov12, Yolov11 and RF-DETR. We will use a dataset of 278 annotated images to train, validate and test the model. We will compare the mAP and precision of the model. The objective of this paper is to identify the model that is best suited for container damage detection. The result is mixed. mAP@50 score of Yolov11 and 12 was 81.9% compared to RF-DETR, which was 77.7%. However, while testing the model for not-so-common damaged containers, the RF-DETR model outperformed the others overall, exhibiting superiority to accurately detecting both damaged containers as well as damage occurrences with high confidence.