Guided Depth Estimation – We present a robust learning framework for extracting feature-based semantic information from a large corpus, with the aim of improving the robustness of state-of-the-art semantic retrieval systems. We evaluate the proposed framework on both large-scale and small-scale datasets, and demonstrate improved performance on a variety of tasks.

In this paper, we present a novel framework for learning 3D models in deep neural network. The proposed framework is based on a deep hierarchical model which consists of hierarchical components and a global topology representation. A deep hierarchical model is designed to learn the model parameters in a deep hierarchy. Then, the model parameters are learned using an embedding procedure. The embedding procedure can be used to dynamically embed parts of the model parameters into the global topology representation. In order to learn the model parameters, the global topology representation and their embedding are jointly learned in a fully supervised manner. We also propose a simple method to learn the model parameters, which utilizes the embedding procedure to learn the model parameters directly from the global topology representation. The proposed deep hierarchical model is shown to learn 3D model parameters efficiently by a real-world problem.

On the Complexity of Bipartite Reinforcement Learning

Towards a Unified Framework for 3D Model Refinement

Guided Depth Estimation

Approximation Algorithms for the Logarithmic Solution of Linear Energies

Deep Multi-Scale Multi-Task Learning via Low-rank Representation of 3D Part FramesIn this paper, we present a novel framework for learning 3D models in deep neural network. The proposed framework is based on a deep hierarchical model which consists of hierarchical components and a global topology representation. A deep hierarchical model is designed to learn the model parameters in a deep hierarchy. Then, the model parameters are learned using an embedding procedure. The embedding procedure can be used to dynamically embed parts of the model parameters into the global topology representation. In order to learn the model parameters, the global topology representation and their embedding are jointly learned in a fully supervised manner. We also propose a simple method to learn the model parameters, which utilizes the embedding procedure to learn the model parameters directly from the global topology representation. The proposed deep hierarchical model is shown to learn 3D model parameters efficiently by a real-world problem.