Transfer Learning Approach for Multilabel Fine Grain Image Attribute Extraction

In the era of artificial intelligence, industries are taking big leap towards transforming overall process of product manufacturing, business measures, retail & customer experience. AI is contributing to small as well as big businesses to enable the automation throughout the pipeline. For steps like product onboarding, attribute identification, PDP enrichment, catalogue generation. Small & mid-level eCommerce industries, in early stage, tends to adopt a low-cost solution for transforming the services. The Literature provides unique and affordable deep learning based multilabel classification approach fore extracting fine grained image level multiple attributes. Converting the objective into multilabel classification problem and using existing stable deep learning models to further train some fine level features, transfer learning is one of the affordable steps to consider. It provides efficient architecture with optimum results, validated on best resources. Extracting product image level fine grained properties is a complex task which requires dense networks with high computing resources. Existing state of the art architectures for classification problems, after tweaking few layers, used as a baseline models. These architectures have proven one of the best solutions for multilabel classification problems with minimal resources required.