VISUAL ATTRIBUTE EXTRACTION OF FASHION IMAGES ON ONLINE SHOP SITES

• Main Author: Aurelio Noviandri, Krishna
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/54161
• Institution: Institut Teknologi Bandung

Online shop sites are currently very popular and widely used. Everyone who has access to the internet can buy and sell various product through an online shop site. One of the most commonly sold and bought product through an online shop is fashion products. Fashion products have various type of attributes. Because all internet users can sell fashion product through an online shop, the descriptions of these fashion product usually are still incomplete and have human error. This can affect the features that are available on the online shop such as search features. Therefore, a solution that can complete the description of fashion product automatically is needed. This solution uses the image of fashion product to extract attributes. The solution model consists of a basic network and several specialized branch network to classify a certain type of attribute. This model use landmarks and bounding box to extract local feature. Local feature extraction for each branch differs according to the type of attribute classified by that branch. The Results shows that the solution model has a progressive nature so the model could follow the growth of online shop site. The use of landmarks and bounding box for local feature extraction only increase a certain type of attributes, where the overall model performance has not improved. Further research is needed to improve local feature extraction used in the model.