Zihao Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Zihao Liu has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Tingting Jiang
National Engineering Research Center of Visual Technology, State Key Laboratory of Multimedia Information Processing, School of Computer Science, Peking University, Beijing, China
4 shared publications - Ruiqin Xiong
Peking University
3 shared publications - Zhiqiang Hu
SenseTime Research
3 shared publications - Qing Xia
SenseTime Research, Beijing, China
2 shared publications - Shaoting Zhang
SenseTime Research, Shanghai, China
2 shared publications - Dimitris N. Metaxas
Rutgers University, New Brunswick, NJ, USA
1 shared publication
Latest Publications
- Open Access
- 13 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. PP, 1
LLMSwitchBench: A New Edge-Cloud Routing Benchmark for Smart Home LLM Inference
In smart home automation, large language models (LLMs) offer transformative potential for enabling intelligent, context-aware interactions. However, deploying LLMs on edge devices is constrained to small-scale LLMs (typically 1B–8B parameters) with limited performance, while cloud-based models, though powerful, raise privacy, latency, and cost concerns. To effectively balance these trade-offs, an intelligent routing strategy is needed to determine when to use local inference and when to offload queries to the cloud. Despite growing attention to this problem, there remains a lack of standardized and comprehensive benchmarks that reflect realistic smart home use cases and support the evaluation of such routing strategies. To address this gap, we propose LLMSwitch-bench, a novelQAbenchmark dataset comprising 6,000 queries across five smart home domains, including dietary, sleep, home security, pet care, and plant care. The dataset includes a wide range of question types with varying levels of difficulty, while been human-validated to ensure accuracy and relevance. Building on this benchmark, we introduce LLMSwitch-router, a privacy-preserving routing method that dynamically balances local and cloud-based inference. LLMSwitch-router prioritizes on-device inference to reduce cost and protect user privacy, while selectively leveraging cloud LLMs for handling complex queries that exceed local LLM capabilities. Experimental results show that LLMSwitch-router achieves a 48% reduction in LLM API cost while maintaining cloud-level accuracy, effectively optimizing the trade-off among performance, cost, and privacy.
- 13 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. PP, 1
Open AccessLLMSwitchBench: A New Edge-Cloud Routing Benchmark for Smart Home LLM Inference
In smart home automation, large language models (LLMs) offer transformative potential for enabling intelligent, context-aware interactions. However, deploying LLMs on edge devices is constrained to small-scale LLMs (typically 1B–8B parameters) with limited performance, while cloud-based models, though powerful, raise privacy, latency, and cost concerns. To effectively balance these trade-offs, an intelligent routing strategy is needed to determine when to use local inference and when to offload queries to the cloud. Despite growing attention to this problem, there remains a lack of standardized and comprehensive benchmarks that reflect realistic smart home use cases and support the evaluation of such routing strategies. To address this gap, we propose LLMSwitch-bench, a novelQAbenchmark dataset comprising 6,000 queries across five smart home domains, including dietary, sleep, home security, pet care, and plant care. The dataset includes a wide range of question types with varying levels of difficulty, while been human-validated to ensure accuracy and relevance. Building on this benchmark, we introduce LLMSwitch-router, a privacy-preserving routing method that dynamically balances local and cloud-based inference. LLMSwitch-router prioritizes on-device inference to reduce cost and protect user privacy, while selectively leveraging cloud LLMs for handling complex queries that exceed local LLM capabilities. Experimental results show that LLMSwitch-router achieves a 48% reduction in LLM API cost while maintaining cloud-level accuracy, effectively optimizing the trade-off among performance, cost, and privacy.
- 16 October 2025
- Published by Elsevier in Neurocomputing
- Vol. 659
DermClinical: Clinical-oriented dataset and evaluation for computer-aided dermatological diagnosis
- 16 October 2025
- Published by Elsevier in Neurocomputing
- Vol. 659
DermClinical: Clinical-oriented dataset and evaluation for computer-aided dermatological diagnosis
- 25 June 2025
- Published by Elsevier in Journal of Investigative Dermatology
- Vol. 146 (1) , 55-63.e7
A 3-Dimensional-Optimized Artificial Imaging Model for the Skin Tumor Burden Assessment of Mycosis Fungoides
- 25 June 2025
- Published by Elsevier in Journal of Investigative Dermatology
- Vol. 146 (1) , 55-63.e7
A 3-Dimensional-Optimized Artificial Imaging Model for the Skin Tumor Burden Assessment of Mycosis Fungoides
- 25 October 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 42 (3) , 619-632
CI-Net: Clinical-Inspired Network for Automated Skin Lesion Recognition
The lesion recognition of dermoscopy images is significant for automated skin cancer diagnosis. Most of the existing methods ignore the medical perspective, which is crucial since this task requires a large amount of medical knowledge. A few methods are designed according to medical knowledge, but they ignore to be fully in line with doctors’ entire learning and diagnosis process, since certain strategies and steps of those are conducted in practice for doctors. Thus, we put forward Clinical-Inspired Network (CI-Net) to involve the learning strategy and diagnosis process of doctors, as for a better analysis. The diagnostic process contains three main steps: the zoom step, the observe step and the compare step. To simulate these, we introduce three corresponding modules: a lesion area attention module, a feature extraction module and a lesion feature attention module. To simulate the distinguish strategy, which is commonly used by doctors, we introduce a distinguish module. We evaluate our proposed CI-Net on six challenging datasets, including ISIC 2016, ISIC 2017, ISIC 2018, ISIC 2019, ISIC 2020 and PH2 datasets, and the results indicate that CI-Net outperforms existing work. The code is publicly available at https://github.com/lzh19961031/Dermoscopy_classification.
- 25 October 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 42 (3) , 619-632
CI-Net: Clinical-Inspired Network for Automated Skin Lesion Recognition
The lesion recognition of dermoscopy images is significant for automated skin cancer diagnosis. Most of the existing methods ignore the medical perspective, which is crucial since this task requires a large amount of medical knowledge. A few methods are designed according to medical knowledge, but they ignore to be fully in line with doctors’ entire learning and diagnosis process, since certain strategies and steps of those are conducted in practice for doctors. Thus, we put forward Clinical-Inspired Network (CI-Net) to involve the learning strategy and diagnosis process of doctors, as for a better analysis. The diagnostic process contains three main steps: the zoom step, the observe step and the compare step. To simulate these, we introduce three corresponding modules: a lesion area attention module, a feature extraction module and a lesion feature attention module. To simulate the distinguish strategy, which is commonly used by doctors, we introduce a distinguish module. We evaluate our proposed CI-Net on six challenging datasets, including ISIC 2016, ISIC 2017, ISIC 2018, ISIC 2019, ISIC 2020 and PH2 datasets, and the results indicate that CI-Net outperforms existing work. The code is publicly available at https://github.com/lzh19961031/Dermoscopy_classification.
- 29 June 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 41 (11) , 3398-3410
Contrastive and Selective Hidden Embeddings for Medical Image Segmentation
Medical image segmentation is fundamental and essential for the analysis of medical images. Although prevalent success has been achieved by convolutional neural networks (CNN), challenges are encountered in the domain of medical image analysis by two aspects: 1) lack of discriminative features to handle similar textures of distinct structures and 2) lack of selective features for potential blurred boundaries in medical images. In this paper, we extend the concept of contrastive learning (CL) to the segmentation task to learn more discriminative representation. Specifically, we propose a novel patch-dragsaw contrastive regularization (PDCR) to perform patch-level tugging and repulsing. In addition, a new structure, namely uncertainty-aware feature re- weighting block (UAFR), is designed to address the potential high uncertainty regions in the feature maps and serves as a better feature re- weighting. Our proposed method achieves state-of-the-art results across 8 public datasets from 6 domains. Besides, the method also demonstrates robustness in the limited-data scenario. The code is publicly available at https://github.com/lzh19961031/PDCR_UAFR-MIShttps://github.com/lzh19961031/PDCR_UAFR-MIS .
- 29 June 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 41 (11) , 3398-3410
Contrastive and Selective Hidden Embeddings for Medical Image Segmentation
Medical image segmentation is fundamental and essential for the analysis of medical images. Although prevalent success has been achieved by convolutional neural networks (CNN), challenges are encountered in the domain of medical image analysis by two aspects: 1) lack of discriminative features to handle similar textures of distinct structures and 2) lack of selective features for potential blurred boundaries in medical images. In this paper, we extend the concept of contrastive learning (CL) to the segmentation task to learn more discriminative representation. Specifically, we propose a novel patch-dragsaw contrastive regularization (PDCR) to perform patch-level tugging and repulsing. In addition, a new structure, namely uncertainty-aware feature re- weighting block (UAFR), is designed to address the potential high uncertainty regions in the feature maps and serves as a better feature re- weighting. Our proposed method achieves state-of-the-art results across 8 public datasets from 6 domains. Besides, the method also demonstrates robustness in the limited-data scenario. The code is publicly available at https://github.com/lzh19961031/PDCR_UAFR-MIShttps://github.com/lzh19961031/PDCR_UAFR-MIS .
- Open Access
Contrastive and Selective Hidden Embeddings for Medical Image Segmentation
Medical image segmentation has been widely recognized as a pivot procedure for clinical diagnosis, analysis, and treatment planning. However, the laborious and expensive annotation process lags down the speed of further advances. Contrastive learning-based weight pre-training provides an alternative by leveraging unlabeled data to learn a good representation. In this paper, we investigate how contrastive learning benefits the general supervised medical segmentation tasks. To this end, patch-dragsaw contrastive regularization (PDCR) is proposed to perform patch-level tugging and repulsing with the extent controlled by a continuous affinity score. And a new structure dubbed uncertainty-aware feature selection block (UAFS) is designed to perform the feature selection process, which can handle the learning target shift caused by minority features with high uncertainty. By plugging the proposed 2 modules into the existing segmentation architecture, we achieve state-of-the-art results across 8 public datasets from 6 domains. Newly designed modules further decrease the amount of training data to a quarter while achieving comparable, if not better, performances. From this perspective, we take the opposite direction of the original self/un-supervised contrastive learning by further excavating information contained within the label.
Open AccessContrastive and Selective Hidden Embeddings for Medical Image Segmentation
Medical image segmentation has been widely recognized as a pivot procedure for clinical diagnosis, analysis, and treatment planning. However, the laborious and expensive annotation process lags down the speed of further advances. Contrastive learning-based weight pre-training provides an alternative by leveraging unlabeled data to learn a good representation. In this paper, we investigate how contrastive learning benefits the general supervised medical segmentation tasks. To this end, patch-dragsaw contrastive regularization (PDCR) is proposed to perform patch-level tugging and repulsing with the extent controlled by a continuous affinity score. And a new structure dubbed uncertainty-aware feature selection block (UAFS) is designed to perform the feature selection process, which can handle the learning target shift caused by minority features with high uncertainty. By plugging the proposed 2 modules into the existing segmentation architecture, we achieve state-of-the-art results across 8 public datasets from 6 domains. Newly designed modules further decrease the amount of training data to a quarter while achieving comparable, if not better, performances. From this perspective, we take the opposite direction of the original self/un-supervised contrastive learning by further excavating information contained within the label.