Chunhui Liu

Chunhui Liu

Auto-generated profile, based on publicly available information

Chunhui Liu

School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu, China

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 January 2021 in Wireless Communications and Mobile Computing
Verified
Journal Article
Open access
PDF file
SHShaoqi Hou
Shaoqi Hou
CLChunhui Liu
Chunhui Liu
KYKangning Yin
Kangning Yin
YDYiyin Ding
Yiyin Ding
ZWZhiguo Wang
Zhiguo Wang
GYGuangqiang Yin
Guangqiang Yin
Person Re‐identification (Re‐ID) is aimed at solving the matching problem of the same pedestrian at a different time and in different places. Due to the cross‐device condition, the appearance of different pedestrians may have a high degree of similarity; at this time, using the global features of pedestrians to match often cannot achieve good results. In order to solve these problems, we designed a Spatial Attention Network Guided by Attribute Label (SAN‐GAL), which is a dual‐trace network containing both attribute classification and Re‐ID. Different from the previous approach of simply adding a branch of attribute binary classification network, our SAN‐GAL is mainly divided into two connecting steps. First, with attribute labels as guidance, we generate Attribute Attention Heat map (AAH) through Grad‐CAM algorithm to accurately locate fine‐grained attribute areas of pedestrians. Then, the Attribute Spatial Attention Module (ASAM) is constructed according to the AHH which is taken as the prior knowledge and introduced into the Re‐ID network to assist in the discrimination of the Re‐ID task. In particular, our SAN‐GAL network can integrate the local attribute information and global ID information of pedestrians without introducing additional attribute region annotation, which has good flexibility and adaptability. The test results on Market1501 and DukeMTMC‐reID show that our SAN‐GAL can achieve good results and can achieve 85.8% Rank‐1 accuracy on DukeMTMC‐reID dataset, which is obviously competitive compared with most Re‐ID algorithms.

24 February 2020 in arXiv
Verified
Preprint
Open access
PDF file
YLYe Li
Ye Li
GYGuangqiang Yin
Guangqiang Yin
CLChunhui Liu
Chunhui Liu
XYxiaoyu yang
xiaoyu yang
ZWZhiguo Wang
Zhiguo Wang
Mining the shared features of same identity in different scene, and the unique features of different identity in same scene, are most significant challenges in the field of person re-identification (ReID). Online Instance Matching (OIM) loss function and Triplet loss function are main methods for person ReID. Unfortunately, both of them have drawbacks. OIM loss treats all samples equally and puts no emphasis on hard samples. Triplet loss processes batch construction in a complicated and fussy way and converges slowly. For these problems, we propose a Triplet Online Instance Matching (TOIM) loss function, which lays emphasis on the hard samples and improves the accuracy of person ReID effectively. It combines the advantages of OIM loss and Triplet loss and simplifies the process of batch construction, which leads to a more rapid convergence. It can be trained on-line when handle the joint detection and identification task. To validate our loss function, we collect and annotate a large-scale benchmark dataset (UESTC-PR) based on images taken from surveillance cameras, which contains 499 identities and 60,437 images. We evaluated our proposed loss function on Duke, Marker-1501 and UESTC-PR using ResNet-50, and the result shows that our proposed loss function outperforms the baseline methods by a maximum of 21.7%, including Softmax loss, OIM loss and Triplet loss.

29 July 2021 in arXiv
Verified
Preprint
Open access
PDF file
CLChunhui Liu
Chunhui Liu
XLXinyu Li
Xinyu Li
HCHao Chen
Hao Chen
DMDavide Modolo
Davide Modolo
JTJoseph Tighe
Joseph Tighe
Most action recognition solutions rely on dense sampling to precisely cover the informative temporal clip. Extensively searching temporal region is expensive for a real-world application. In this work, we focus on improving the inference efficiency of current action recognition backbones on trimmed videos, and illustrate that one action model can also cover then informative region by dropping non-informative features. We present Selective Feature Compression (SFC), an action recognition inference strategy that greatly increase model inference efficiency without any accuracy compromise. Differently from previous works that compress kernel sizes and decrease the channel dimension, we propose to compress feature flow at spatio-temporal dimension without changing any backbone parameters. Our experiments on Kinetics-400, UCF101 and ActivityNet show that SFC is able to reduce inference speed by 6-7x and memory usage by 5-6x compared with the commonly used 30 crops dense sampling procedure, while also slightly improving Top1 Accuracy. We thoroughly quantitatively and qualitatively evaluate SFC and all its components and show how does SFC learn to attend to important video regions and to drop temporal features that are uninformative for the task of action recognition.