SELECTED PUBLICATIONS

• propose dynamic weight slicing scheme, achieving good hardware-efficiency by predictively slicing network parameters at test time with respect to different inputs.

Changlin Li, Guangrun Wang, Xiaodan Liang, Zhihui Li, and Xiaojun Chang*

IEEE Trans. Pattern Anal. Mach. Intell. (T-PAMI), DOI:10.1109/TPAMI.2022.3194044, 2022

pdf

• We propose a novel framework TN-ZSTAD for zero-shot temporal activity detection, which combines a graph activity transformer (AGT) and a zeroshot detection subnet (ZSDN) together to infer a set of unseen activity instances for a input video directly.

Lingling Zhang, Xiaojun Chang, Jun Liu, Minnan Luo, Zhihui Li, Lina Yao, and Alex Hauptmann

IEEE Trans. Pattern Anal. Mach. Intell. (T-PAMI), DOI:10.1109/TPAMI.2022.3183586, 2022

pdf

• We propose a novel method to leverage visual contents to synthesize additional pseudo-pivoting for unsupervised multimodal machine translation.

Mingjie Li, Poyao Huang, Xiaojun Chang*, Junjie Hu, Yi Yang and Alex Hauptmann

IEEE Trans. Pattern Anal. Mach. Intell. (T-PAMI), DOI:10.1109/TPAMI.2022.3181116, 2022

pdf

• We develop a novel semantics-guided contrastive net- work for ZSD, underpinned by a new mapping- contrastive strategy superior to the conventional mapping-transfer strategy. To the best of our knowl- edge, this is the first work that applies contrastive learning mechanism for ZSD.

Caixia Yan, Xiaojun Chang*, Minnan Luo, Huan Liu, Xiaoqin Zhang*, and Qinghua Zheng

IEEE Trans. Pattern Anal. Mach. Intell. (T-PAMI), DOI:10.1109/TPAMI.2021.3140070, 2021

pdf

• Considering the varieties in datasets and tasks, we make the first attempt to bring NAS techniques into the realm of ZSL, and thus propose ZeroNAS to formulate the GAN architecture design for ZSL as a NAS problem.

Caixia Yan, Xiaojun Chang*, Zhihui Li, Weili Guan, Zongyuan Ge, Lei Zhu and Qinghua Zheng

IEEE Trans. Pattern Anal. Mach. Intell. (T-PAMI), DOI:10.1109/TPAMI.2021.3127346, 2021

pdf | code

• This survey conducts a comprehensive investigation of the current scene graph research

Xiaojun Chang, Pengzhen Ren, Pengfei Xu, Zhihui Li, Xiaojiang Chen and Alex Hauptmann

IEEE Trans. Pattern Anal. Mach. Intell. (T-PAMI), DOI:10.1109/TPAMI.2021.3137605, 2021

pdf

• To improve transferability, we further devised a variant of NSAS, called NSAS-C, which searches for deeper architectures in the convolutional cell search.

Miao Zhang, Huiqi Li, Shirui Pan, Xiaojun Chang, Chuan Zhou, Zongyuan Ge, and Steven Su

IEEE Trans. Pattern Anal. Mach. Intell. (T-PAMI), 2020

pdf | code

• design the informed nearly-isotonic SVM classifier (NI-SVM) that is able to exploit the carefully constructed ordering information

Xiaojun Chang, Yaoliang Yu, Yi Yang, Eric P. Xing

IEEE Trans. Pattern Anal. Mach. Intell. (T-PAMI) 39(8) 1617-1632 (2017)

pdf | code

• we provide a new perspective, beginning with an overview of the characteristics of the earliest NAS algorithms, summarizing the problems in these early NAS algorithms, and then providing solutions for subsequent related research work

Pengzhen Ren, Yun Xiao, Xiaojun Chang, Po-Yao Huang, Zhihui Li, Xiaojiang Chen and Xin Wang

Accepted by ACM Computing Surveys, 2021

pdf

• propose a semi-supervised batch mode multi-class active learning algorithm for visual concept recognition

Yi Yang, Zhigang Ma, Feiping Nie, Xiaojun Chang, Alexander G. Hauptmann

International Journal of Computer Vision (IJCV) 113(2) 113-127 (2015)

pdf

• propose a semantics preserving graph propagation model based on GCN for the challenging ZSD task, which leverages both the semantic description and structural knowledge in prior category graphs to facilitate the semantic coherency of region graph.

Caixia Yan, Qinghua Zheng, Xiaojun Chang, Minnan Luo, ChungHsing Yeh and Alexander G. Hauptmann

IEEE Transactions on Image Processing 29:8163-8176 (2020)

pdf

• formulate a multi-cycle consistency loss based selfsupervised learning manner to pre-training the deep feature extraction network, which can take advantage of extensive unlabeled video samples rather than limited manually annotated samples

Di Yuan, Xiaojun Chang, Po-Yao Huang, Qiao Liu, and Zhenyu He

IEEE Transactions on Image Processing 30:976-985 (2021)

pdf

• we present an effective cross-modal clinical graph transformer for ophthalmic report generation

M. Li, W. Cai, K. Verspoor, S. Pan, X. Liang and X. Chang*

CVPR 2022

pdf

• propose a novel plug-and-play module with a dynamic multi-scale window for multi-head selfattention in transformer

P. Ren, C. Li, G. Wang, Y. Xiao, Q. Du, X. Liang, and X. Chang

CVPR 2022

pdf

• propose a global scaling-up-down prediction method and a local 3D-2D-3D projectionreconstruction method for point cloud domain adaptation

H. Fan, X. Chang, W. Zhang, Y. Cheng, Y. Sun, and M. Kankanhalli

CVPR 2022

pdf

• develop a strong manual baseline for progressive learning of ViTs, by introducing MoGrow, a momentum growth strategy to bridge the gap brought by model growing

C. Li, B. Zhuang, G. Wang, X. Liang, X. Chang, and Y. Yang

CVPR 2022

pdf

• this paper formulates the neural architecture search as a distribution learning problem through relaxing the architecture weights into Gaussian distributions

M. Zhang, S. Pan, X. Chang, S. Su, J. Hu, R. Haffari, and B. Yang

CVPR 2022

pdf

• propose a distinct Dual-path Actor Interaction (DualAI) framework, which flexibly arranges spatial and temporal transformers in two complementary orders, enhancing actor relations by integrating merits from different spatiotemporal paths

M. Han, D. J. Zhang, Y. Wang, R. Yan, L. Yao, X. Chang*, and Y. Qiao

CVPR 2022

pdf

• propose a novel one-to-all spatial matching knowledge distillation approach

S. Lin, H. Xie, B. Wang, K, Yu, X. Chang, X. Liang, and G. Wang

CVPR 2022

pdf

• present a new benchmark, FFA-IR, towards an explainable and reliable MRG benchmark based on FFA Images and Reports

M. Li, W. Cai, R. Liu, Y. Weng, X. Zhao, C. Wang, X. Chen, Z. Liu, C. Pan, M. Li, Y. Zheng, Y. Liu, F. D. Salim, K. Verspoor, X. Liang, X. Chang*

NeurIPS 2021

openreview

• introduce the inter-channel correlation, with the characteristics of being invariant to the spatial dimen- sion, to explore and measure both the feature diversity and homology to help the student for better represen- tation learning

L. Liu, Q. Huang, S. Lin, H. Xie, B. Wang, X. Chang* and X. Liang

ICCV 2021

arxiv | pdf | poster

• propose the Random Environmen- tal Mixup (REM) method, which generates cross-connected house scenes as augmented data via mixuping environment

C. Liu, F. Zhu, X. Chang, X. Liang, Z. Ge and Y. Shen

ICCV 2021

arxiv | pdf

• present Block-wisely Self-supervised Neural Architecture Search (BossNAS), an unsupervised NAS method that addresses the problem of in- accurate architecture rating caused by large weight-sharing space and biased supervision in previous methods

C. Li, T. Tang, G. Wang, J. Peng, B. Wang, X. Liang and X. Chang

ICCV 2021

arxiv | pdf | CODE

• propose a new dynamic network routing regime, achieving good hardware-efficiency by predictively adjusting filter numbers of networks at test time with respect to different inputs.

C. Li, G. Wang, B. Wang, X. Liang, Z. Li, and X. Chang

CVPR 2021

paper | ORAL | CODE

• propose a task named Scenario Oriented Object Navigation (SOON), in which an agent is instructed to find an object in a house from an arbitrary starting position

F. Zhu, X. Liang, Y. Zhu, X. Chang, Q. Yu, and X. Liang

CVPR 2021

paper

• propose a universal policy decoupling transformer model that extends MARL to a much broader scenario

S. Hu, F. Zhu, X. Chang and X. Liang

ICLR 2021

paper

• This paper deepens our understanding of the hypergradient calculation in the differentiable NAS.

M. Zhang, S. Su, S. Pan, X. Chang, E. Abbasnejad and R. Haffari

ICML 2021

paper

• design a novel Inter-Video Proposal Relation method, which can effectively leverage inter-video proposal relation to learn discriminative representations for video object detection

M. Han, Y. Wang, X. Chang and Y. Qiao

ECCV 2020

paper | code

• investigate how to utilize visual content for disambiguation and latent space alignment in unsupervised MMT

P. Huang, J. Hu, X. Chang and A. Hauptmann

ACL 2020

paper

• Introducing Auxiliary Reasoning Navigation (AuxRN), a framework with four self-supervised auxiliary reasoning tasks to take advantage of the additional training signals derived from the semantic information

F. Zhu, Y. Zhu, X. Chang, X. Liang

CVPR 2020

paper | DEMO | CODE | Oral

• Proposing a novel problem setting for temporal activity detection in which activities that are not seen during the training stage can be recognized and localized simultaneously

L. Zhang, X. Chang, J. Liu, S. Wang, Z. Ge, M. Luo, A. Hauptmann

CVPR 2020

paper

• smooth the style disparities within the same camera and across different cameras

C. Liu, X. Chang, Y. Shen

CVPR 2020

paper

• modularize the large search space of NAS into blocks to ensure that the potential candidate architectures are fully trained

C. Li, J. Peng, L. Yuan, G. Wang, X. Liang, L. Lin, X. Chang

CVPR 2020

paper | code

• learning an agent endowed with the capability of constant conversation for help with natural language and navigating according to human responses
• propose a Cross-modal Memory Network (CMN) for remembering and understanding the rich information relevant to historical navigation actions

Y. Zhu, F. Zhu, Z. Zhan, B. Lin, J. Jiao, X. Chang, X. Liang

CVPR 2020

paper | code

• formulate the supernet training in the One-Shot NAS as a constrained optimization problem of continual learning that the learning of current architecture should not degrade the performance of previous architectures during the supernet training

M. Zhang, H. Li, S. Pan, X. Chang, S. Su

CVPR 2020

paper | code

• enhance the intelligent exploration of differentiable Neural Architecture Search in the latent space

M. Zhang, H. Li, S. Pan, X. Chang, Z. Ge and S. Su

NeurIPS 2020

paper | code

• leverage the volumetric stereo matching pipeline and allow the network to automat70 ically select the optimal structures for both the Feature Net and the Matching Net

X. Cheng, Y. Zhong, M. Harandi, Y. Dai, X. Chang, H. Li, T. Drummond, and Z. Ge

NeurIPS 2020

paper | code

• leveraging visual object detection and propose a model with diverse multi-head attention to learn grounded multilingual multimodal representations

P. Huang, X. Chang, A. G. Hauptmann

EMNLP 2019

paper

• propose a novel framework to leverage automatically extracted regional semantics from un-annotated images as additional weak supervision to learn visual-semantic embeddings

P. Huang, G. Kang, W. Liu, X. Chang, A. G. Hauptmann

ACM MM 2019

paper

• made the earliest attempt to address the person search problem and built the first deep reinforcement learning based person search framework

X. Chang, P. Huang, Y. Shen, X. Liang, Y. Yang and A. G. Hauptmann

ECCV 2018

paper

• make a pioneer effort to formulate the video object segmentation problem as a Markov Decision Process and propose a novel reinforcement cutting-agent learning framework to tackle this problem

J. Han, L. Yang, D. Zhang, X. Chang, X. Liang

CVPR 2018

paper

• simultaneously learns a linear SVM classifier and infers a binary indicator for each instance in order to select reliable training instances from each positive or negative bag

H. Fan, X. Chang, D. Cheng, Y. Yang, D. Xu, A. G. Hauptmann

ICCV 2017

paper

• combine the concept classifiers based on a principled estimate of their accuracy on the unlabeled test videos

X. Chang, Y. Yu, Y. Yang, E. P. Xing

CVPR 2016

paper

• define a novel notion of semantic saliency that assesses the relevance of each shot with the event of interest

X. Chang, Y. Yang, E. P. Xing, Y. Yu

ICML 2015

paper | code

• propose a joint framework that simultaneously detects high-level events and localizes the indicative concepts of the events

X. Chang, Y. Yu, Y. Yang, A. G. Hauptmann

ACM MM 2015

paper

• we propose Multi-granularity Auxiliary Reason- ing Navigation (MG-AuxRN) to facilitate navigation learning. MG-AuxRN perceives multi-granularity input which combining dense object features and global image features.

Fengda Zhu, Yi Zhu, Yanxin Long, Xiaojun Chang, and Xiaodan Liang

Submitted to IEEE Trans. Pattern Anal. Mach. Intell. (T-PAMI), 2021

pdf | code