Adobe Launched Artificial Intelligence Fitting Technology, Even The Wrinkles Of Clothes Can Be Clearly Presented.

A group of researchers from Adobe's media and data science research laboratory, Indian Institutet of Technology and Stanford University developed an image-based virtual try technology, named SieveNet, which can map a garment to a virtual human body, thus retaining the characteristics of a garment (including folds) without causing blurred or coloured colors. This technology will help e-commerce platform operators improve product photos on the website, and can also satisfy consumers' trial loading needs when shopping in physical stores.
In recent years, with the development of the electricity supplier industry, many businesses have begun to enter the electronic business platform. In online shopping, consumers can not intuitively feel the quality of products, and can only use product photos on the website for reference. However, in order to save time and resources, many businesses will simply copy and paste many products onto the model to get the product effect diagram quickly, so that consumers can not distinguish the material and form of clothing products. In view of this problem, in recent years, virtual assembly technology has been widely researched and applied in the clothing industry.
Adobe said that the newly developed SieveNet system is equipped with advanced neural network to perfect the characteristics of a garment when it is trying, such as the material of the clothes and the fitting effect of the different shapes of the models (even the wrinkles of the clothes can be fully reflected). At the same time, the quality of the picture will not be affected.
VentureBeat describes the principle of SieveNet system in a document, which is based on clothing images and human body models, and generates new images of the models wearing the clothing (retaining the original body shape, posture and other details). For this reason, it adopts multi-stage technology, including changing the garment to make it consistent with the posture and shape of the mannequin, and then transferring the textured texture to the model.
The researchers used 19 thousand sets of images of positive female models and the images of top products to train SieveNet systems, which were inconsistent in body characteristics. In qualitative tests, the system is better than the benchmark in dealing with problems such as occlusion, postural change, color leakage and geometric warping. In addition, it has achieved the most advanced results in qualitative indicators, including the Fr e Chet Inception Distance (FID) index, which can get photos from the target distribution and the evaluated system (SieveNet), and use AI object recognition system to capture important features and retain similarity. Adobe researchers say that one of the advantages of SieveNet is that it can be compatible with most existing websites and mobile app.
SieveNet is not the first system to develop a virtual fitting function. Previously, the US / India artificial intelligence platform start-up company Vue.ai used its image and video recognition technology to transform simple clothing images into new images of clothing displayed by models. The generated images provide consumers with more practical visual effects, and also save money for retailers. This automatic imaging technology is 5 times faster than traditional photography. Vue.ai completed its $17 million B round of financing last year.
3DLook, a human body scanning startups, uses smart phones to scan the body and get body parameters to help consumers match the most fitting clothes. Users can also perform virtual try on smart phones and observe the effect of trying. Just one cell phone and two photos, 3DLook can help users complete body shape measurement and scanning. The company won the LVMH LVMH third Award last year (Innovation).
Brands like Gucci and Nike have developed a mobile phone app for consumers to test shoes.
The beauty industry is taking the lead in the application of virtual reality technology. France's beauty giant L "Or" Al (L'OREAL)'s AR technology developer ModiFace has so far developed hair, beauty and jewellery AR applications for more than 80 brands Amazon, Sephora and Estee Lauder. Similar technology providers also include Perfect's YouCam, which allows 3D scanning of the face to achieve virtual make-up of lips, eyes, eyebrows, hair and cheeks. In addition, there are also AR beauty nail ManiMatch and China beauty chart AR beauty makeup camera MakeupPlus.