Ji Zhang

I am Ji Zhang (张继), an Assistant Professor in the School of Computing and Artificial Intelligence at Southwest Jiaotong University (SWJTU). I received my Ph.D. degree in 2024 from the School of Computer Science and Engineering, University of Electronic Science and Technology of China (UESTC), where I was very fortunate to be advised by Prof. Jingkuan Song and Prof. Lianli Gao.

My research interests include Robotics, Computer Vision and Few-shot Learning. I am particularly interested in designing advanced algorithms that exploit robotic foundation models (i.e., vision-language-action models (VLAs)) to tackle challenging real-world robotic applications. If you are also interested in related topics, please do not hesitate to reach out.

Scholar / Github / Email / WeChat / 中文主页

News

[04/2025] I have been invited to be a PC member for NeurIPS'25.
[03/2025] I have been invited to be the PC members for ICCV'25 and ACM MM'25.
[02/2025] Our paper about vision-language models (VLMs) was accepted by CVPR'25.
[11/2024] I have been invited to be a PC member for CVPR'25.
[08/2024] I have been invited to be a PC member for ICLR'25.
[02/2024] Our paper about vision-language models (VLMs) was accepted by CVPR'24.
[10/2023] Our paper about out-of-distirbution (OOD) detection was accepted by IEEE TIP.
[10/2023] I'm awarded the "Graduate Chinese National Scholarship".
[07/2023] Our paper about few-shot learning was accepted by ICCV'23.
[04/2023] Our paper about few-shot learning was accepted by ICML'23.
[06/2022] Our papers about few-shot learning and continue learning are accepted by ACM MM'22.
[03/2022] Our paper about meta-learning was accepted by IEEE TCSVT.
[06/2021] Our paper about meta-learning was accepted by ACM MM'21.

Selected Papers

(* Equal contribution. # Corresponding author)

	InSpire: Vision-Language-Action Models with Intrinsic Spatial Reasoning Ji Zhang, Shihan Wu, Xu Luo, Hao Wu, Lianli Gao, Heng Tao Shen, Jingkuan Song arXiv, 2025 [Paper][Project] Mitigating the adverse effects of spurious correlations by boosting the spatial reasoning ability of vision-language-action (VLA) models.
	Policy Contrastive Decoding for Robotic Foundation Models Shihan Wu, Ji Zhang**, Xu Luo, Junlin Xie, Lianli Gao, Heng Tao Shen, Jingkuan Song arXiv, 2025 [Paper][Project] A simple, training-free, easy-to-implement and plug-and-play scheme for addressing the spurious correlation issue in robot policies.
	Skip Tuning: Pre-trained Vision-Language Models are Effective and Efficient Adapters Themselves Shihan Wu, Ji Zhang#, Pengpeng Zeng, Lianli Gao, Jingkuan Song, Heng Tao Shen IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2025 [Paper][Code] Achieving effective and efficient adaptation of large pre-trained vision-language models.
	DePT: Decoupled Prompt Tuning Ji Zhang, Shihan Wu, Lianli Gao, Heng Tao Shen, Jingkuan Song IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2024 [Paper][Code] Overcoming the base-new tradeofff (BNT) problem for existing prompt tuning methods.
	From Global to Local: Multi-scale Out-of-distribution Detection Ji Zhang, Lianli Gao, Bingguang Hao, Hao Huang, Jingkuan Song, Heng Tao Shen IEEE Transactions on Image Procesing (TIP), 2023 [Paper][Code] Leveraging both global visual information and local region details of images to maximally benefit OOD detection.
	DETA: Denoised Task Adaptation for Few-shot Learning Ji Zhang, Lianli Gao, Xu Luo, Heng Tao Shen, Jingkuan Song IEEE International Conference on Computer Vision (ICCV), 2023 [Paper][Code] Tacking both the X-noise (i.e., image noise) and the Y-noise (i.e., label noise) in a unified framework for test-time few-shot tasks.
	A Closer Look at Few-shot Classification Again Xu Luo, Hao Wu, Ji Zhang, Lianli Gao, Jing Xu, Jingkuan Song International Conference on Machine Learning (ICML), 2023 [Paper][Code] Empirically proving the disentanglement of training and test-time adaptation algorithms in few-shot classification.
	Free-lunch for Cross-domain Few-shot learning: Style-aware Episodic Training with Robust Contrastive Learning Ji Zhang, Jingkuan Song, Lianli Gao, Heng Tao Shen ACM International Conference on Multimedia (ACM MM), 2022 [Paper][Code] Addressing the side-effect of style-shift between tasks from source and target domains.
	Progressive Meta-learning with Curriculum Ji Zhang, Jingkuan Song, Lianli Gao, Ye Liu, Heng Tao Shen IEEE Transactions on Circuits and Systems for Video Technology (TCSVT), 2022 [Paper][Code] An extended version the ACM MM'21 paper, where the curriculum is effectively integrated as a regularization term into the objective so that the meta-learner can measure the hardness of tasks adaptively.
	Curriculum-based Meta-learning Ji Zhang, Jingkuan Song, Yazhou Yao, Lianli Gao ACM International Conference on Multimedia (ACM MM), 2021 [Paper][Code] Progressively improving the meta-learner by performing episodic training on simulating tasks from easy to hard, i.e., in a curriculum learning manner.

Academic Service

I'm a reviewer of several top journals/conferences, e.g. CVPR, ICCV, NeurIPS, ICLR, ACM MM, AAAI, IEEE TPAMI, IEEE TIP.

This well-designed template is borrowed from Jonbarron.