Amazon industrialise ce que font des startups
Fashion e-commerce sites often show photos of clothing in two ways: the clothing when worn by models, or shown folded / hung without the model. For the second of these, professional photo editors are employed to create ‘packshot’ photos that show the clothing in a way that simulates it being worn by a human. This is costly and time consuming. Especially for e-commerce sites that have multiple suppliers where the photos are less likely to be standardised – e.g. different lighting conditions, different photo qualities, different poses shown with the clothing.
Amazon is working on automating packshot photos from input images of models wearing clothing. Using machine learning, Amazon will segment the clothing from the image of the model, and then normalize it according to a certain specification. For instance, the clothing could be shown as if it was worn with the same model poses, without any visible differences in lighting, and without any obstructions (e.g. a model’s hand blocking an item).
Moreover, Amazon could use generative adversarial networks (a form of AI) to ‘create’ models for each piece of clothing. This might be useful for creating standardised images of model-worn clothing, especially if an e-commerce platform has multiple suppliers.
In the context of Amazon’s business, it has multiple suppliers who are offering clothing, but a big constraint to the success of this category is the quality (and its variance) of the photos taken by the suppliers. By using its segmentation technology, Amazon could firstly isolate the clothing items and standardize how they’re presented, and then secondly create new model images that are consistently high quality. This could lift the sales of this overall category for Amazon, as well as providing a valuable, cost & time saving service to suppliers.
via Patent Drop : lire l’article source