Object detection has made impressive progress in recent years with the help of deep learning. However, state-of-the-art algorithms are both computation and memory intensive. Though many lightweight networks are developed for a trade-off between accuracy and efficiency, it is still a challenge to make it practical on an embedded device. In this paper, we present a system-level solution for efficient object detection on a heterogeneous embedded device. The detection network is quantized to low bits and allows efficient implementation with shift operators. In order to make the most of the benefits of low-bit quantization, we design a dedicated accelerator with programmable logic. Inside the accelerator, a hybrid dataflow is exploited according to the heterogeneous property of different convolutional layers. We adopt a straightforward but resource-friendly column-prior tiling strategy to map the computation-intensive convolutional layers to the accelerator that can support arbitrary feature size. Other operations can be performed on the low-power CPU cores, and the entire system is executed in a pipelined manner. As a case study, we evaluate our object detection system on a real-world surveillance video with input size of 512x512, and it turns out that the system can achieve an inference speed of 18 fps at the cost of 6.9W (with display) with an mAP of 66.4 verified on the PASCAL VOC 2012 dataset.
Long-range dependencies modeling, widely used in capturing spatiotemporal correlation, has shown to be effective in CNN dominated computer vision tasks. Yet neither stacks of convolutional operations to enlarge receptive fields nor recent nonlocal modules is computationally efficient. In this paper, we present a generic family of lightweight global descriptors for modeling the interactions between positions across different dimensions (e.g., channels, frames). This descriptor enables subsequent convolutions to access the informative global features with negligible computational complexity and parameters. Benchmark experiments show that the proposed method can complete state-of-the-art long-range mechanisms with a significant reduction in extra computing cost. Code available at https://github.com/HolmesShuan/Compact-Global-Descriptor.
Existing person re-identification has achieved great progress in the visible domain, capturing all the person images with visible cameras. However, in a 24-hour intelligent surveillance system, the visible cameras may be noneffective at night. In this situation, thermal cameras are the best supplemental components, which capture images without depending on visible light. Therefore, in this paper, we investigate the visible-thermal cross-modality person re-identification (VT Re-ID) problem. In VT Re-ID, there are two knotty problems should be well handled, cross-modality discrepancy and intra-modality variations. To address these two issues, we propose focusing on enhancing the discriminative feature learning (EDFL) with two extreme simple means from two core aspects, (1) skip-connection for mid-level features incorporation to improve the person features with more discriminability and robustness, and (2) dual-modality triplet loss to guide the training procedures by simultaneously considering the cross-modality discrepancy and intra-modality variations. Additionally, the two-stream CNN structure is adopted to learn the multi-modality sharable person features. The experimental results on two datasets show that our proposed EDFL approach distinctly outperforms state-of-the-art methods by large margins, demonstrating the effectiveness of our EDFL to enhance the discriminative feature learning for VT Re-ID.
Non-uniform blur, mainly caused by camera shake and motions of multiple objects, is one of the most common causes of image quality degradation. However, the traditional blind deblurring methods based on blur kernel estimation do not perform well on complicated non-uniform motion blurs. Recent studies show that GAN-based approaches achieve impressive performance on deblurring tasks. In this letter, to further improve the performance of GAN-based methods on deblurring tasks, we propose an edge heuristic multi-scale generative adversarial network(GAN), which uses the "coarse-to-fine" scheme to restore clear images in an end-to-end manner. In particular, an edge-enhanced network is designed to generate sharp edges as auxiliary information to guide the deblurring process. Furthermore, We propose a hierarchical content loss function for deblurring tasks. Extensive experiments on different datasets show that our method achieves state-of-the-art performance in dynamic scene deblurring.
Deep metric learning aims at learning the distance metric between pair of samples, through the deep neural networks to extract the semantic feature embeddings where similar samples are close to each other while dissimilar samples are farther apart. A large amount of loss functions based on pair distances have been presented in the literature for guiding the training of deep metric learning. In this paper, we unify them in a general pair-based weighting loss function, where the minimizing objective loss is just the distances weighting of informative pairs. The general pair-based weighting loss includes two main aspects, (1) samples mining and (2) pairs weighting. Samples mining aims at selecting the informative positive and negative pair sets to exploit the structured relationship of samples in a mini-batch and also reduce the number of non-trivial pairs. Pair weighting aims at assigning different weights for different pairs according to the pair distances for discriminatively training the network. We detailedly review those existing pair-based losses inline with our general loss function, and explore some possible methods from the perspective of samples mining and pairs weighting. Finally, extensive experiments on three image retrieval datasets show that our general pair-based weighting loss obtains new state-of-the-art performance, demonstrating the effectiveness of the pair-based samples mining and pairs weighting for deep metric learning.
In this paper, we focus on model generalization and adaptation for cross-domain person re-identification (Re-ID). Unlike existing cross-domain Re-ID methods, leveraging the auxiliary information of those unlabeled target-domain data, we aim at enhancing the model generalization and adaptation by discriminative feature learning, and directly exploiting a pre-trained model to new domains (datasets) without any utilization of the information from target domains. To address the discriminative feature learning problem, we surprisingly find that simply introducing the attention mechanism to adaptively extract the person features for every domain is of great effectiveness. We adopt two popular type of attention mechanisms, long-range dependency based attention and direct generation based attention. Both of them can perform the attention via spatial or channel dimensions alone, even the combination of spatial and channel dimensions. The outline of different attentions are well illustrated. Moreover, we also incorporate the attention results into the final output of model through skip-connection to improve the features with both high and middle level semantic visual information. In the manner of directly exploiting a pre-trained model to new domains, the attention incorporation method truly could enhance the model generalization and adaptation to perform the cross-domain person Re-ID. We conduct extensive experiments between three large datasets, Market-1501, DukeMTMC-reID and MSMT17. Surprisingly, introducing only attention can achieve state-of-the-art performance, even much better than those cross-domain Re-ID methods utilizing auxiliary information from the target domain.
Semi-supervised learning (SSL) provides a powerful framework for leveraging unlabeled data when labels are limited or expensive to obtain. SSL algorithms based on deep neural networks have recently proven successful on standard benchmark tasks. However, many of them have thus far been either inflexible, inefficient or non-scalable. This paper explores recently developed contrastive predictive coding technique to improve discriminative power of deep learning models when a large portion of labels are absent. Two models, cpc-SSL and a class conditional variant~(ccpc-SSL) are presented. They effectively exploit the unlabeled data by extracting shared information between different parts of the (high-dimensional) data. The proposed approaches are inductive, and scale well to very large datasets like ImageNet, making them good candidates in real-world large scale applications.
Temporal action detection aims at not only recognizing action category but also detecting start time and end time for each action instance in an untrimmed video. The key challenge of this task is to accurately classify the action and determine the temporal boundaries of each action instance. In temporal action detection benchmark: THUMOS 2014, large variations exist in the same action category while many similarities exist in different action categories, which always limit the performance of temporal action detection. To address this problem, we propose to use joint Identification-Verification network to reduce the intra-action variations and enlarge inter-action differences. The joint Identification-Verification network is a siamese network based on 3D ConvNets, which can simultaneously predict the action categories and the similarity scores for the input pairs of video proposal segments. Extensive experimental results on the challenging THUMOS 2014 dataset demonstrate the effectiveness of our proposed method compared to the existing state-of-art methods for temporal action detection in untrimmed videos.