Knowledge distillation is an effective technique that has been widely used for transferring knowledge from a network to another network. Despite its effective improvement of network performance, the dependence of accompanying assistive models complicates the training process of single network in the need of large memory and time cost. In this paper, we design a more elegant self-distillation mechanism to transfer knowledge between different distorted versions of same training data without the reliance on accompanying models. Specifically, the potential capacity of single network is excavated by learning consistent global feature distributions and posterior distributions (class probabilities) across these distorted versions of data. Extensive experiments on multiple datasets (i.e., CIFAR-10/100 and ImageNet) demonstrate that the proposed method can effectively improve the generalization performance of various network architectures (such as AlexNet, ResNet, Wide ResNet, and DenseNet), outperform existing distillation methods with little extra training efforts.

Channel pruning is an effective technique that has been widely applied to deep neural network compression. However, many existing methods prune from a pretrained model, thus resulting in repetitious pruning and fine-tuning processes. In this article, we propose a dynamical channel pruning method, which prunes unimportant channels at the early stage of training. Rather than utilizing some indirect criteria (e.g., weight norm, absolute weight sum, and reconstruction error) to guide connection or channel pruning, we design criteria directly related to the final accuracy of a network to evaluate the importance of each channel. Specifically, a channelwise gate is designed to randomly enable or disable each channel so that the conditional accuracy changes (CACs) can be estimated under the condition of each channel disabled. Practically, we construct two effective and efficient criteria to dynamically estimate CAC at each iteration of training; thus, unimportant channels can be gradually pruned during the training process. Finally, extensive experiments on multiple data sets (i.e., ImageNet, CIFAR, and MNIST) with various networks (i.e., ResNet, VGG, and MLP) demonstrate that the proposed method effectively reduces the parameters and computations of baseline network while yielding the higher or competitive accuracy. Interestingly, if we Double the initial Channels and then Prune Half (DCPH) of them to baseline's counterpart, it can enjoy a remarkable performance improvement by shaping a more desirable structure.

To harvest small networks with high accuracies, most existing methods mainly utilize compression techniques such as low-rank decomposition and pruning to compress a trained large model into a small network or transfer knowledge from a powerful large model (teacher) to a small network (student). Despite their success in generating small models of high performance, the dependence of accompanying assistive models complicates the training process and increases memory and time cost. In this article, we propose an elegant self-distillation (SD) mechanism to obtain high-accuracy models directly without going through an assistive model. Inspired by the invariant recognition in the human vision system, different distorted instances of the same input should possess similar high-level data representations. Thus, we can learn data representation invariance between different distorted versions of the same sample. Especially, in our learning algorithm based on SD, the single network utilizes the maximum mean discrepancy metric to learn the global feature consistency and the Kullback-Leibler divergence to constrain the posterior class probability consistency across the different distorted branches. Extensive experiments on MNIST, CIFAR-10/100, and ImageNet data sets demonstrate that the proposed method can effectively reduce the generalization error for various network architectures, such as AlexNet, VGGNet, ResNet, Wide ResNet, and DenseNet, and outperform existing model distillation methods with little extra training efforts.

Monocular 6D pose estimation for objects is an essential but challenging task that is commonly applied in computer vision and robotics. Existing two-stage methods solve for rotations with Perspective-n-Point (PnP), which still incorporates translations, resulting in accuracy degeneration. In contrast, direct regression methods adopt Convolutional Neural Networks (CNNs) to solve for rotations and translations jointly but suffer from performance gaps in rotation accuracy. In this article, we propose a novel Normal Vector guided Regression Network (NVR-Net) to directly regress the 6D pose from a single RGB image under the guidance of 3D normal vectors. Specifically, we design a novel Orientation-Aware Feature (OAF) for pose estimation. It consists of two corresponding sets of 3D normal vectors to thoroughly disentangle rotation from translation estimation. Then, we introduce a CNN to predict a dense pixelwise representation of the OAF without viewpoint ambiguity. To estimate rotations and translations individually from the OAF, we propose a novel Pose from Normal Vectors (PNV) head networks under the instruction of a differentiable closed-form solution. Finally, extensive experiments on three common benchmarks demonstrate that our approach outperforms state-of-the-art methods on rotation accuracy and removes the gap between indirect and end-to-end methods. Moreover, our method can estimate the 6D pose of a single object within an RGB image in real-time.

This article addresses the challenge of 6D aircraft pose estimation from a single RGB image during the flight. Many recent works have shown that keypoints-based approaches, which first detect keypoints and then estimate the 6D pose, achieve remarkable performance. However, it is hard to locate the keypoints precisely in complex weather scenes. In this article, we propose a novel approach, called Pose Estimation with Keypoints and Structures (PEKS), which leverages multiple intermediate representations to estimate the 6D pose. Unlike previous works, our approach simultaneously locates keypoints and structures to recover the pose parameter of aircraft through a Perspective-n-Point Structure (PnPS) algorithm. These representations integrate the local geometric information of the object and the topological relationship between components of the target, which effectively improve the accuracy and robustness of 6D pose estimation. In addition, we contribute a dataset for aircraft pose estimation which consists of 3681 real images and 216,000 rendered images. Extensive experiments on our own aircraft pose dataset and multiple open-access pose datasets (e.g., ObjectNet3D, LineMOD) demonstrate that our proposed method can accurately estimate 6D aircraft pose in various complex weather scenes while achieving the comparative performance with the state-of-the-art pose estimation methods.