In recent years, automatic check-out (ACO) gains increasing interest and has been widely used in daily life. However, current works mainly rely on both counter and product prototype images in the training phase, and it is hard to maintain the performance in an incremental setting. To deal with this problem, in this paper, we propose a robust prototype-f ree r etrieval method (ROPREM) for ACO, which is a cascaded framework composed of a product detector module and a product retrieval module. Specifically, we use the product detector module to locate the products and then deliver the results to the subsequent product retrieval module for counting. The product detector module is trained without product class information, which can avoid the model overfitting for the known classes and improve the performance for the incremental setting. Additionally, we treat the check-out process as a retrieval process rather than a classification process. The retrieval result is considered as the product class by calculating the feature similarity between the query and gallery templates. The quantities of each category are treated as the final counts. To our best knowledge, this is the first attempt to treat ACO as a retrieval task. Extensive experiments are conducted to validate the proposed ROPREM, and the results show that ROPREM achieves the best performance in comparison with several state-of-the-art methods on the public retail product checkout (RPC) dataset.
