CLDM

Due to the huge market related with clothing, the clothing fashion analysis is receiving more and more attentions from both industry and academia. Because of the shooting conditions and the human pose in the clothing image, the clothing items may be distorted or occluded, which makes the clothing fashion applications inaccurate. In this paper, we present CLDM, a clothing landmark detector based on Mask R-CNN. CLDM detects the functional regions of cloth items, which makes the features extracted from clothes more discriminative. We propose the architecture of CLDM, and design the deep learning network, during which RestNet-FPN is adopted as the backbone and a landmark branch is designed in the perception network. We implement CLDM and experimentally validate the accuracy and efficiency of our proposed techniques.