PaintNet:A Shape-Constrained Generative Framework For Clothing Generation From Fashion Model

Recent years have witnessed the proliferation of e-commerce platforms such as Taobao and JD.com,where clothing trading is especially flourishing.However,the large number and the diverse kinds of clothing may cause people to be lost in the enormous product sea,and unable to pick up their favorite clothing products.To address this problem,researchers have carried out a lot of studies in the field of fashion,and the research towards clothing generation from fashion model images has been ascendant.The research towards clothing generation from fashion model images can facilitate many downstream applications,which has a high commercial value.In fact,clothing generation from fashion model is essentially to accomplish domain transfer.Compared with traditional domain transfer tasks,clothing generation from fashion model has larger domain gap,thus the task of clothing generation from fashion model images faces much more difficult challenges.To address the above challenges,we propose a two-stage shape-constrained clothing generative framework,dubbed as PaintNet.PaintNet comprises two coherent components:shape predictor and texture renderer.The shape predictor is devised to predict the shape map based on representation learning and nearest neighbor strategy.At the same time,the texture renderer is introduced to generate the final clothing image with the guidance of the predicted shape map and segmented texture.We adopt the Lookbook dataset to conduct extensive experiments to verify the effectiveness of PaintNet.The experimental results show that PaintNet achieves the best performance.Moreover,we also explore the potential of PaintNet in the task of crossdomain clothing retrieval,and the experimental results show that PaintNet can achieve,on average,5.34%performance improvement over the traditional non-generative retrieval methods.