Relative Depth Order Estimation Using Multi-Scale Densely Connected Convolutional Networks

Abstract

We study the problem of estimating the relative depth order of point pairs in a monocular image. Recent advances [1], [2] mainly focus on using deep convolutional neural networks (DCNNs) to learn and infer the ordinal information from multiple contextual information of the points pair such as global scene context, local contextual information, and the locations. However, it remains unclear how much each context contributes to the task. To address this, we first examine the contribution of each context cue [1], [2] to the performance in the context of depth order estimation. We find out the local context surrounding the points pair contributes the most and the global scene context helps little. Based on the findings, we propose a simple method, using a multi-scale densely-connected network to tackle the task. Instead of learning the global structure, we dedicate to explore the local structure by learning to regress from regions of multiple sizes around the point pairs. Moreover, we use the recent densely connected network [3] to encourage substantial feature reuse as well as deepen our network to boost the performance. We show in experiments that the results of our approach is on par with or better than the state-of-the-art methods with the benefit of using only a small number of training data.

Fusion Learning for sEMG Recognition of Multiple Upper-Limb Rehabilitation Movements

“…Meanwhile, in order to prevent too large or too small a range of features from affecting the convergence speed of the network, the normalization layers and activation layers are added after the network input layer [ 43 , 44 , 45 ]. Subsequently, the time–frequency features of the sEMG signals are concatenated and sent to the transition layer, which controls the feature map’s size and considers the compatibility of the feature map with the computation burden of the network [ 46 ]. For the purpose of acquiring more comprehensive features about rehabilitation actions [ 47 ], the above operations are performed repeatedly to gain the deeper features of sEMG signals.…”

Underwater Communication Signal Recognition Using Sequence Convolutional Network

“…In Fig. 11, the employed network compared with the typical solutions is as follow: LSTM [54] and GRU [55] are the familiar form of RNN; ResNet [56], DensenNet [57], SENet [58] are the sophisticated structure of deep residual neural network with more layer-by-layer connections; PnasNet [59] is the irregular structure generated by reinforcement learning method; Deep neural network (DNN) [29], extensible neural network (ENN) [32], Sparse Autoencoder [35] are in the light of the regular one-dimensional fully connected layer to construct the network; the improved CNN [30], CLDNN (Convolutional Long Short-term Deep Neural Network) [60], fusion neural network (FNN) [34] are made up of CNN and RNN; K Nearest Neighbors [13], Random Forest [15], support vector machines (SVM) [14] are frequently used ML methods in AMR work. All networks are trained and verified at packets = 2187.…”