We propose that non-rigid 3D structures are well modeled by a dictionary sparsely, a more expressive and general assumption. Based on this basic assumption we present two approaches without the aid of additional priors (e.g. temporal ordering, rigid substructures, etc.) to solve non-rigid structure from motion problem.
We investigate the problem of estimating the dense 3D shape of an object, given a set of 2D landmarks and silhouette in a single image. An obvious prior to employ in such a problem is myriad of dense CAD models online. As each model is manually and independently designed and does not necessarily share the same number of vertices or the same structure of meshes. We propose a novel graph embedding based on local dense correspondence to allow for sparse linear combinations of CAD models.
We analyze convolutional neural network via convolutional sparse coding. We propose to simplify the CNN architecture by replacing non-unity stride by unity stride. We employ a novel alternation strategy for CNN training that leads to substantially faster convergence rates, nice theoretical properties, and achieving state of the art results across large scale datasets (e.g. ImageNet) as well as other standard benchmarks.