59-05-032 Proceeding

79 Proceedings of the Princess Maha Chakri Sirindhorn Congress interconnect. NVLink will enable very dense compute farms where a single node can have 8 or more GPUs. IV. Fashion Product Search and Results In this paper, we present the research at the ROSE Lab on visual search for fashion related products, specifically, on handbag, shoes and clothes. We will also demonstrate their correlated applications and demos for illustration. A. Handbag For those companies well-known for selling luxury handbags, one of the main issues is how tomake their handbags more popular to their customers. This will require the manufacturers to gain a deep insight into the customers’ feedback about their handbags. A traditional way to get the consumers’ comments about one particular handbag is to do a keyword based search through the internet. However, nowadays people prefer to upload the photos of their purchases on blogs or twitters without describing the specific names or models, which brings difficulties for manufacturers to collect users’ feedback. Therefore, it would be necessary to develop an image based handbag recognition engine for these manufacturers. In order to identify the subtle differences among handbags, we propose to enhance the handbag local structure pattern, and extract the feature from the enhanced handbag image to complement the feature extracted directly from the original handbag image. We term these two types of features as the complementary and original features. We also test different deep learning features for handbag recognition. 1) Branded Handbag Dataset : We build a branded handbag dataset for one particular brand: Louis Vuitton. Images in our dataset are downloaded fromGoogle, Flicker as well as some online shopping websites. Eventually, the dataset contains 220 handbag models (classes). We randomly partition our handbag dataset into two parts, five images per handbag for training and the rest for testing. Figure 1 shows the examples handbag images in the database. Figure 1 Examples of handbag images in the database