Weakly supervised barcodes detection

Barcodes allow us to encode a various amount of useful information. It is important to quickly and accurately find their type and location in correct recognition images. In this paper, we propose an approach of barcodes detection based on neural networks using weakly labeled data. Using this approach, we can find barcodes and even classify them. The proposed approach does not require the exact location of objects in the markup, which greatly simplifies the process of obtaining data to train a neural network model. The proposed approach shows a high quality of barcode detection in images, viz. 0,725 precision, 0,674 recall, 0,698 F1.