Course Project
Retrieving product information from real-world image can have many promising applications. We propose a cross domain generative adversarial model to address the challenge of generating clothes image from a dressed person. We train the neural network with paired images of fashion models(source domain) and products (target domain). To improve generated image quality and similarity, we introduce an additional association discriminator into original Generative adversarial network (GAN) model during training. We further enhance network performance by replacing convolutional layers with residual blocks. Experimental results show that we succeed in extracting clothes image from photos with great quality.