Performance Evaluation of Container Identification Detection Algorithm

The identification of container surfaces carries a large amount of crucial information regarding production and logistics. Research on the detection and identification of containers is lacking in both academia and industry, and the efficiency is low due to the need for manual completion of related tasks. In order to tackle this problem, we have created a large-scale text detection dataset for container surface identification called IdentificationText. This dataset consists of 12,000 high-resolution images, providing bounding boxes annotations for text detection tasks. We have discussed downstream applications of the IdentificationText dataset as well as our annotation techniques used in the dataset’s creation. The text in this dataset exhibits challenges such as deformations, multi-direction, and multi-scale. We conducted extensive experiments to evaluate the effectiveness and difficulty of this dataset using advanced text detection methods. In our experiments, we found that repeated textures and vertical text at multiple scales would cause missed detections, which was an extremely serious problem. The experimental results indicate that it is challenging for current text detection methods to locating text on container surfaces. Achieving higher accuracy in detecting text on containers requires more in-depth research. The experimental results serve as the benchmark performance for the IdentificationText dataset, providing reference for future researchers.