WorldCache: Accelerating World Models for Free via Heterogeneous Token Caching
Weilun Feng†, Guoxin Fan†, Haotong Qin†, Mingqiang Wu, Yuqi Li, Xiangqi Li, Zhulin An*, Libo Huang, Dingrui Wang, Longlong Liao, Michele Magno, Yongjun Xu, Chuanguang Yang*.
in International Conference on Machine Learning (ICML-2026)
CCF-A, Acceptance rate: 6352/23918=26.6%
[Paper] [Code]
We propose WorldCache, a training-free heterogeneous token caching framework that accelerates diffusion-based world models by prioritizing hard, rapidly changing tokens while reusing predictable ones.

Fast-SAM3D: 3Dfy Anything in Images but Faster
Weilun Feng†, Mingqiang Wu†, Zhiliang Chen, Chuanguang Yang*, Haotong Qin, Yuqi Li, Xiaokun Liu, Guoxin Fan, Libo Huang, Yulun Zhang, Michele Magno, Yongjun Xu, Zhulin An*.
in International Conference on Machine Learning (ICML-2026)
CCF-A, Acceptance rate: 6352/23918=26.6%
[Paper] [Code]
We propose Fast-SAM3D, a training-free framework that accelerates single-view 3D generation through modality-aware caching, spatiotemporal token carving, and spectral-aware token aggregation.

Distilling Out-of-Distribution Knowledge from Large Language Models for CLIP Generalization
Qiji Ma, Chuanguang Yang*, Zhulin An*, Libo Huang, Erhu Zhao, Yuqi Li, Yongjun Xu.
in IEEE/CVF Conference on Computer Vision and Pattern Recognition Findings (CVPRF-2026)
[Paper] [BibTex]
We enhance prompt-based CLIP generalization by distilling out-of-distribution knowledge from LLM-generated descriptions into synthesized visual features for both base and novel classes.

EfficientVPR: Toward Efficient Visual Place Recognition via Scene-Aware Prompt Tuning and Adaptive Feature Enhancement
Wenjing Tang, Chuanguang Yang*, Zhulin An*, Libo Huang, Boyu Diao, Yongjun Xu.
in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR-2026)
CCF-A, Acceptance rate: 4089/16092=25.4%
[Paper] [Code] [BibTex]
We propose EfficientVPR, a lightweight one-stage visual place recognition framework combining scene-aware prompt tuning with instance-dependent local feature enhancement.

QuantSparse: Comprehensively Compressing Video Diffusion Transformer with Model Quantization and Attention Sparsification
Weilun Feng, Chuanguang Yang*, Haotong Qin, Mingqiang Wu, Yuqi Li, Xiangqi Li, Zhulin An*, Libo Huang, Yulun Zhang, Michele Magno, Yongjun Xu.
in International Conference on Learning Representations (ICLR-2026)
CCF-A, Acceptance rate: 5355/19525=27.4%
[Paper] [Code] [BibTex]
We propose QuantSparse, a unified video diffusion Transformer compression framework that combines model quantization and attention sparsification through multi-scale attention distillation and second-order sparse attention reparameterization.

Quantized Visual Geometry Grounded Transformer
Weilun Feng, Haotong Qin, Mingqiang Wu, Chuanguang Yang*, Yuqi Li, Xiangqi Li, Zhulin An*, Libo Huang, Yulun Zhang, Michele Magno, Yongjun Xu.
in International Conference on Learning Representations (ICLR-2026)
CCF-A, Acceptance rate: 5355/19525=27.4%
[Paper] [BibTex]
We propose QuantVGGT, a post-training quantization framework tailored to Visual Geometry Grounded Transformers that reduces memory and computation while preserving 3D reconstruction accuracy.

S²Q-VDiT: Accurate Quantized Video Diffusion Transformer with Salient Data and Sparse Token Distillation
Weilun Feng†, Haotong Qin†, Chuanguang Yang*, Xiangqi Li, Han Yang, Yuqi Li, Zhulin An*, Libo Huang, Michele Magno, Yongjun Xu.
in Advances in Neural Information Processing Systems (NeurIPS-2025)
CCF-A, Acceptance rate: 5290/21575=24.5%
[Paper] [Code]
We propose Hessian-aware Salient Data Selection and Attention-guided Sparse Token Distillation for accurate quantized video diffusion transformer.

TCMP-300: A Comprehensive Traditional Chinese Medicinal Plant Dataset for Plant Recognition
Yanling Zhang, Wanhui Sun, Chuanguang Yang*, Libo Huang*, Zhulin An*, Weilun Feng, Wenjing Tang, Yongjun Xu.
in Scientific Data-2025
Nature branch journal
[Paper] [Code] [BibTex]
We present a Traditional Chinese Medicinal Plant (TCMP) dataset that includes 52,089 images in 300 categories to promote the development and validation of advanced AI models for robust and accurate TCMP recognition.

Frequency-Aligned Knowledge Distillation for Lightweight Spatiotemporal Forecasting
Yuqi Li†, Chuanguang Yang†, Hansheng Zeng, Zeyu Dong, Zhulin An, Yongjun Xu, Yingli Tian, Hao Wu*.
in IEEE/CVF International Conference on Computer Vision (ICCV-2025)
CCF-A, Acceptance rate: 2698/11239=24%
[Paper] [Code]
We propose a frequency-aligned knowledge distillation strategy, which extracts multi-scale spectral features from the teacher's latent space, including both high and low frequency components, to guide the lightweight student model in capturing both local finegrained variations and global evolution patterns.

Merlin: Multi-View Representation Learning for Robust Multivariate Time Series Forecasting with Unfixed Missing Rates
Chengqing Yu, Fei Wang*, Chuanguang Yang, Zezhi Shao, Tao Sun, Tangwen Qian, Wei Wei, Zhulin An, Yongjun Xu*.
in ACM SIGKDD Conference on Knowledge Discovery and Data Mining (SIGKDD-2025)
CCF-A
[Paper] [Code]
We propose Multi-View Representation Learning, including offline knowledge distillation and multi-view contrastive learning, for robust multivariate time series forecasting with unfixed missing rates.

Q-VDiT: Towards Accurate Quantization and Distillation of Video-Generation Diffusion Transformers
Weilun Feng, Chuanguang Yang*, Haotong Qin, Xiangqi Li, Yu Wang, Zhulin An*, Libo Huang, Boyu Diao, Zixiang Zhao, Yongjun Xu, Michele Magno.
in International Conference on Machine Learning (ICML-2025)
CCF-A, Acceptance rate: 3260/12107=26.9%
[Paper] [Code] [BibTex]
We propose Token-aware Quantization Estimator and Temporal Maintenance Distillation for video DiT models.

Geometric Feature Embedding for Effective 3D Few-Shot Class Incremental Learning
Xiangqi Li†, Libo Huang†, Zhulin An*, Weilun Feng, Chuanguang Yang, Boyu Diao, Fei Wang, Yongjun Xu.
in International Conference on Machine Learning (ICML-2025)
CCF-A, Acceptance rate: 3260/12107=26.9%
[Paper] [Code] [BibTex]
We propose a geometric feature embedding module by augmenting text prompts with spatial geometric features for effective 3D few-shot class incremental learning.

Multi-party Collaborative Attention Control for Image Customization
Han Yang, Chuanguang Yang*, Qiuli Wang, Zhulin An*, Weilun Feng, Libo Huang, Yongjun Xu.
in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR-2025)
CCF-A, Acceptance rate: 2878/13008=22.1%
[Paper] [Code] [BibTex]
We propose Multi-party Collaborative Attention Control (MCA-Ctrl), a tuning-free method that enables high-quality image customization using both text and complex visual conditions.

Multi-Teacher Knowledge Distillation with Reinforcement Learning for Visual Recognition
Chuanguang Yang, Xinqiang Yu, Han Yang, Zhulin An*, Chengqing Yu, Libo Huang, Yongjun Xu.
in AAAI Conference on Artificial Intelligence (AAAI-2025 Oral)
CCF-A, Acceptance rate: 3032/12957=23.40%
Oral rate: 600/12957=4.63%
[Paper] [Code] [BibTex]
We propose Multi-Teacher Knowledge Distillation with Reinforcement Learning (MTKD-RL) to dynamically optimize multi-teacher weights for visual recognition models.

MPQ-DM: Mixed Precision Quantization for Extremely Low Bit Diffusion Models
Weilun Feng†, Haotong Qin†, Chuanguang Yang*, Zhulin An*, Libo Huang, Boyu Diao, Fei Wang, Renshuai Tao, Yongjun Xu, Michele Magno.
in AAAI Conference on Artificial Intelligence (AAAI-2025 Oral)
CCF-A, Acceptance rate: 3032/12957=23.40%
Oral rate: 600/12957=4.63%
[Paper] [Code] [BibTex]
We propose MPQ-DM, a Mixed-Precision Quantization method for extremely low bit diffusion models.

HSRDiff: A Hierarchical Self-Regulation Diffusion Model for Stochastic Semantic Segmentation
Han Yang, Chuanguang Yang*, Zhulin An*, Libo Huang, Yongjun Xu.
in AAAI Conference on Artificial Intelligence (AAAI-2025)
CCF-A, Acceptance rate: 3032/12957=23.40%
[Paper] [Code] [BibTex]
We propose a Hierarchical Self-Regulation Diffusion (HSRDiff) Model for stochastic semantic segmentation.

Relational Diffusion Distillation for Efficient Image Generation
Weilun Feng, Chuanguang Yang, Zhulin An, Libo Huang, Boyu Diao, Fei Wang, Yongjun Xu.
in ACM International Conference on Multimedia (ACM MM-2024 Oral)
CCF-A, Acceptance rate: 1149/4385=26.20%
Oral rate: 174/4385=3.97%
[Paper] [Code] [BibTex]
We propose Relational Diffusion Distillation, a novel distillation method tailored specifically for distilling diffusion models.

DetKDS: Knowledge Distillation Search for Object Detectors
Lujun Li, Yufan Bao, Peijie Dong, Chuanguang Yang, Anggeng Li, Wenhan Luo, Qifeng Liu, Wei Xue, Yike Guo.
in International Conference on Machine Learning (ICML-2024)
CCF-A, Acceptance rate: 2610/9473=27.5%
[Paper] [Code] [BibTex]
We leverage search algorithms to discover optimal distillers for object detectors.

Online Policy Distillation with Decision-Attention
Xinqiang Yu, Chuanguang Yang, Chengqing Yu, Libo Huang, Zhulin An, Yongjun Xu.
in International Joint Conference on Neural Networks (IJCNN-2024)
CCF-C, Acceptance rate: 1701/3272=51.99%
[Paper] [BibTex]
We propose Online Policy Distillation with Decision-Attention to make various policies operate in the same environment learn different perspectives.

CLIP-KD: An Empirical Study of CLIP Model Distillation
Chuanguang Yang, Zhulin An, Libo Huang, Junyu Bi, Xinqiang Yu, Han Yang, Boyu Diao, Yongjun Xu.
in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR-2024)
CCF-A, Acceptance rate: 2719/11532=23.6%
[Paper] [Code] [BibTex]
We propose several distillation strategies, including relation, feature, gradient and contrastive paradigms, to examine the effectiveness of CLIP-Knowledge Distillation.

eTag: Class-Incremental Learning with Embedding Distillation and Task-Oriented Generation
Libo Huang, Yan Zeng, Chuanguang Yang, Zhulin An, Boyu Diao, Yongjun Xu.
in AAAI Conference on Artificial Intelligence (AAAI-2024)
CCF-A, Acceptance rate: 2342/12100=23.8%
[Paper] [Code] [BibTex]
We propose a method of embedding distillation and task-oriented generation for class-incremental learning, which requires neither the exemplar nor the prototype.

VL-Match: Enhancing Vision-Language Pretraining with Token-Level and Instance-Level Matching
Junyu Bi, Daixuan Cheng, Ping Yao, Bochen Pang, Yuefeng Zhan, Chuanguang Yang, et al.
in IEEE/CVF International Conference on Computer Vision (ICCV-2023)
CCF-A, Acceptance rate: 2160/8068=26.8%
[Paper] [BibTex]
We propose VL-Match, a Vision-Language framework with Enhanced Token-level and Instance-level Matching.

Categories of Response-Based, Feature-Based, and Relation-Based Knowledge Distillation
Chuanguang Yang, Xinqiang Yu, Zhulin An, Yongjun Xu.
in Advancements in Knowledge Distillation: Towards New Horizons of Intelligent Systems (Springer Book Chapter)
Invited Survey Paper
[Paper] [BibTex]
This paper provides a comprehensive KD survey, including knowledge categories, distillation schemes and algorithms, as well as some empirical studies on performance comparison.

Online Knowledge Distillation via Mutual Contrastive Learning for Visual Recognition
Chuanguang Yang, Zhulin An, Helong Zhou, Fuzhen Zhuang, Yongjun Xu, Qian Zhang.
in IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI-2023)
CCF-A
[Paper] [Code] [BibTex]
We present a Mutual Contrastive Learning (MCL) framework for online KD. The core idea of MCL is to perform mutual interaction and transfer of contrastive distributions among a cohort of networks in an online manner.

MixSKD: Self-Knowledge Distillation from Mixup for Image Recognition
Chuanguang Yang, Zhulin An, Helong Zhou, Linhang Cai, Xiang Zhi, Jiwen Wu, Yongjun Xu, Qian Zhang.
in European Conference on Computer Vision (ECCV-2022)
CCF-B, Acceptance rate: 1650/5803=28.4%
[Paper] [Code] [BibTex]
This paper presents MixSKD, a powerful Self-KD method to regularize the network to behave linearly in feature maps and class probabilities between samples using Mixup images.

Knowledge Distillation Using Hierarchical Self-Supervision Augmented Distribution
Chuanguang Yang, Zhulin An, Linhang Cai, Yongjun Xu
in IEEE Transactions on Neural Networks and Learning Systems (TNNLS-2022)
CCF-B, IF:14.255
[Paper] [Code] [BibTex]
We investigate an auxiliary self-supervision augmented distributionan for training a single network, offline KD and online KD.

Localizing Semantic Patches for Accelerating Image Classification
Chuanguang Yang, Zhulin An, Yongjun Xu
in IEEE International Conference on Multimedia and Expo (ICME-2022)
CCF-B, Acceptance rate: 381/1285=29.6%
[Paper] [Code] [BibTex]
We propose an interpretable AnchorNet to localize semantic patches for accelerating image classification.

Cross-Image Relational Knowledge Distillation for Semantic Segmentation
Chuanguang Yang, Helong Zhou, Zhulin An, Xue Jiang, Yongjun Xu, Qian Zhang
in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR-2022)
CCF-A, Acceptance rate: 2067/8161=25.3%
[Paper] [Code] [BibTex]
We propose cross-Image relational knowledge distillation for semantic segmentation.

Mutual Contrastive Learning for Visual Representation Learning
Chuanguang Yang, Zhulin An, Linhang Cai, Yongjun Xu
in AAAI Conference on Artificial Intelligence (AAAI-2022 Oral)
CCF-A; Acceptance Rate: 1349/9020=15.0%; Oral Rate: Top 5%
[Paper] [Code] [BibTex]
We propose a simple yet effective mutual contrastive learning approach to learn better feature representations for both supervised and self-supervised image classification.

Prior Gradient Mask Guided Pruning-aware Fine-tuning
Linhang Cai, Zhulin An, Chuanguang Yang, Yanchun Yang, Yongjun Xu
in AAAI Conference on Artificial Intelligence (AAAI-2022)
CCF-A; Acceptance Rate: 1349/9020=15.0%
[Paper] [Code] [BibTex]
We proposed a prior gradient mask guided pruning-aware fine-tuning framework to accelerate CNNs for image classification.

Hierarchical Self-supervised Augmented Knowledge Distillation
Chuanguang Yang, Zhulin An, Linhang Cai, Yongjun Xu
in International Joint Conference on Artificial Intelligence (IJCAI-2021)
CCF-A; Acceptance Rate: 587/4204=13.9%
[Paper] [Code] [BibTex]

We propose a strong self-supervised augmented knowledge distillation method from hierarchical feature maps for image classification.

Multi-View Contrastive Learning for Online Knowledge Distillation
Chuanguang Yang, Zhulin An, Yongjun Xu
in IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP-2021)
CCF-B; Acceptance Rate: 1734/3610=48.0%
[Paper] [Code] [BibTex]

We propose multi-view contrastive learning to perform online knowledge distillation for image classification.

DRNet: Dissect and Reconstruct the Convolutional Neural Network via Interpretable Manners
Xiaolong Hu, Zhulin An, Chuanguang Yang, Hui Zhu, Kaiqaing Xu, Yongjun Xu
in European Conference on Artificial Intelligence (ECAI-2020)
CCF-B; Acceptance Rate: 365/1363=26.8%
[Paper] [BibTex]

We propose an interpretable manner to dynamically run channels for sub-class image classification.

Gated Convolutional Networks with Hybrid Connectivity for Image Classification
Chuanguang Yang, Zhulin An, Hui Zhu, Xiaolong Hu, Kun Zhang, Kaiqiang Xu, Chao Li, Yongjun Xu
in AAAI Conference on Artificial Intelligence (AAAI-2020)
CCF-A; Acceptance Rate: 1591/7737=20.6%
[Paper] [Code] [BibTex]

We propose a new network architecture called HCGNet equipped with novel hybrid connectivity and gated mechanisms for image classification.

EENA: Efficient Evolution of Neural Architecture
Hui Zhu, Zhulin An, Chuanguang Yang, Kaiqiang Xu, Erhu Zhao, Yongjun Xu
in International Conference on Computer Vision Workshops (ICCVW-2019)
CCF-A Workshop
[Paper] [BibTex]

We propose an efficient evolution algorithm for neural architecture search for image classification.

Multi-objective Pruning for CNNs using Genetic Algorithm
Chuanguang Yang, Zhulin An, Chao Li, Boyu Diao, Yongjun Xu
in International Conference on Artificial Neural Networks (ICANN-2019)
CCF-C
[Paper] [BibTex]

We propose a genetic algorithm to deal with model pruning using a multi-objective fitness function.