Enhancing Quick and Accurate Static 3D Cloth Draping by Curvature Loss with GarNet++ PROJECT TITLE : GarNet++: Improving Fast and Accurate Static 3D Cloth Draping by Curvature Loss ABSTRACT: In this paper, we address the issue of static cloth draping on virtual human bodies using three-dimensional models. We present a two-stream deep network model that, when applied to virtual 3D bodies, generates a visually plausible draping of a template cloth by extracting features from both the body and the garment shape. Our neural network is trained to simulate a physics-based simulation (PBS) method while simultaneously reducing the amount of computation time needed by two orders of magnitude. In order to train the network, we first introduce loss terms that are based on PBS. These terms help produce results that are plausible and make the model aware of collisions. We introduce two loss functions that penalize the difference between the curvature of the predicted cloth and PBS in order to increase the amount of detail that is seen in the draped garment. In particular, we focus on studying the influence of mean curvature normal and a new detail-preserving loss method from both a qualitative and quantitative standpoint. The local covariance matrices of the three-dimensional points are computed by our newly developed curvature loss, and the Rayleigh quotients of the prediction and PBS are compared with one another. This results in more details while performing favorably or comparably when compared to the loss that takes into consideration the mean curvature normal vectors in the 3D triangulated meshes. Our framework is validated using four different types of clothing for a variety of body types and poses. Finally, our results are significantly better than those obtained by a data-driven method that was just recently proposed. Did you like this research project? To get this research project Guidelines, Training and Code... Click Here facebook twitter google+ linkedin stumble pinterest Toward Concept-based Item Representation Learning with the Item Concept Network Representation Learning for Activity with Multi-level Attention Kinematic Similarity Computation