We study network pruning which aims to remove redundant channels/kernels and hence speed up the inference of deep networks. Existing pruning methods either train from scratch with sparsity constraints or minimize the reconstruction error between the feature maps of the pre-trained models and the compressed ones. Both strategies suffer from some limitations: the former kind is computationally expensive and difficult to converge, while the latter kind optimizes the reconstruction error but ignores the discriminative power of channels. In this paper, we propose a simple-yet-effective method called discrimination-aware channel pruning (DCP) to choose the channels that actually contribute to the discriminative power. To this end, we first introduce additional discrimination-aware losses into the network to increase the discriminative power of the intermediate layers. Next, we select the most discriminative channels for each layer by considering the discrimination-aware loss and the reconstruction error, simultaneously. We then formulate channel pruning as a sparsity-inducing optimization problem with a convex objective and propose a greedy algorithm to solve the resultant problem. Note that a channel (3D tensor) often consists of a set of kernels (each with a 2D matrix). Besides the redundancy in channels, some kernels in a channel may also be redundant and fail to contribute to the discriminative power of the network, resulting in kernel level redundancy. To solve this issue, we propose a discrimination-aware kernel pruning (DKP) method to further compress deep networks by removing redundant kernels. To avoid manually determining the pruning rate for each layer, we propose two adaptive stopping conditions to automatically determine the number of selected channels/kernels. The proposed adaptive stopping conditions tend to yield more efficient models with better performance in practice. Extensive experiments on both image classification and face recognition demonstrate the effectiveness of our methods. For example, on ILSVRC-12, the resultant ResNet-50 model with 30 percent reduction of channels even outperforms the baseline model by 0.36 percent in terms of Top-1 accuracy. We also deploy the pruned models on a smartphone (equipped with a Qualcomm Snapdragon 845 processor). The pruned MobileNetV1 and MobileNetV2 achieve 1.93× and 1.42× inference acceleration on the mobile device, respectively, with negligible performance degradation. The source code and the pre-trained models are available at https://github.com/SCUT-AILab/DCP.

Information processing of guide dog robot requires expensive computing resources to meet real-time performance. We propose an edge computing framework based on Intel Up-Squared board and neural compute stick. Image processing and real-time control are performed on the framework. In addition, the voice recognition and commanding are also implemented, which are processed on Amazon cloud service. Vision, speech, and control services are integrated by ASU VIPLE (Visual IoT/Robotics Programming Language Environment). A prototype is developed, which implements the guide dog's obstacle avoidance, traffic sign recognition and following, and voice interaction with human.

Channel pruning is one of the predominant approaches for deep model compression. Existing pruning methods either train from scratch with sparsity constraints on channels, or minimize the reconstruction error between the pre-trained feature maps and the compressed ones. Both strategies suffer from some limitations: the former kind is computationally expensive and difficult to converge, whilst the latter kind optimizes the reconstruction error but ignores the discriminative power of channels. To overcome these drawbacks, we investigate a simple-yet-effective method, called discrimination-aware channel pruning, to choose those channels that really contribute to discriminative power. To this end, we introduce additional losses into the network to increase the discriminative power of intermediate layers and then select the most discriminative channels for each layer by considering the additional loss and the reconstruction error. Last, we propose a greedy algorithm to conduct channel selection and parameter optimization in an iterative way. Extensive experiments demonstrate the effectiveness of our method. For example, on ILSVRC-12, our pruned ResNet-50 with 30% reduction of channels even outperforms the original model by 0.39% in top-1 accuracy.

Object recognition and modeling have classically been studied separately, but practically, they are two closely correlated aspects. In this paper, by exploring the interrelations, we propose a framework to address these two problems at the same time, which we call simultaneous recognition and modeling. Differing from traditional recognition process which consists of off-line object model learning and on-line recognition procedures, our method is solely online. Starting with an empty object database, we incrementally build up object models while at the same time using these models to identify newly observed object views. In the proposed framework, objects are modeled as view graphs and a probabilistic observation model is presented. Both the appearance and the spatial structure of the object are examined, and a formulation based on maximum likelihood estimation is developed. Joint object recognition and modeling are achieved by solving the optimization problem. To evaluate the framework, we have developed a method for simultaneously learning multiple 3-D object models directly from the cluttered indoor environment and tested it using several everyday scenes. Experimental results demonstrate that the framework can cope with the recognition and modeling problem together nicely.

3D model building is an essential topic in both computer vision and robotic community. Usually this is achieved with alignment methods which try to place the data into a common reference frame by finding out the transformations between them. Traditional local alignment approaches, such as ICP and its variants, mainly concern about pairwise constraints and leave out the transformation consistency requirements implied by data sequence. However, these consistencies are important clues to increase the precision of alignment. To use this information better, we present Local Transformation Consistent Alignment (LTCA), which takes triple of frames as input and simultaneously optimize the transformations between them so that local consistencies can be maintained. To evaluate the proposed method, we apply it to the 3D modeling problem with RGB-D image data as input. Comparisons with the state-of-the-art pairwise alignment- Generalized ICP are also made. Experimental results show that more precise alignment results can be obtained when consistency constraints are kept.