Aiming to reduce online shopping returns, Amazon has developed a novel approach to enhance the online shopping experience for both buyers and sellers.
The company’s latest innovation, “Diffuse to Choose” (DTC), is an AI-powered tool crafted by Amazon researchers. This advanced technology enables customers to virtually try on clothing or visualize how furniture would look in their own spaces. DTC offers real-time insights into how these items would fit and appear, thereby improving decision-making for online shoppers.
DTC represents a cutting-edge approach in the realm of “virtual trial and error” through its diffusion-based image inpainting model. As described in the research paper, this technology excels in blending detailed and contextually accurate visuals with realistic lighting and shadow effects.
DTC employs a unique dual-network structure to enhance image processing for products. The first network manages the general aspects of the image, while the second is dedicated exclusively to the product.
This two-pronged strategy is crucial for preserving the product’s detailed features, like replicating the texture of a sofa’s fabric. Researchers emphasize that DTC outperforms current methodologies, delivering precise results without the necessity for product-specific training.
Amazon’s new AI model has been training on the company’s own datasets as well as publicly available ones.
As with most other modern AI models, this one also comes with its own share of limitations, most of them having to do with its Stable Diffusion base. Some of the problems include incorrect poses for people or suboptimal text rendering. The problem with poses could easily be fixed with more training, but the researchers specifically chose to make the AI tool more broadly applicable rather than focusing on specialized training in specific fields.
Regardless, DTC should still make shopping decisions a lot easier for Amazon users with a more personalized experience. For retailers, this innovation presents an opportunity to foster deeper customer loyalty and potentially reduce return rates, as it allows for a more accurate representation of products before purchase.
Amazon plans to release the code and demo for UTC soon, but there is no precise launch date yet.
Via: The Decoder
