Search

Your search keyword '"Zeng, Zhilin"' showing total 156 results
Start Over Author "Zeng, Zhilin"
156 results on '"Zeng, Zhilin"'

2. Parameter-efficient Fine-tuning in Hyperspherical Space for Open-vocabulary Semantic Segmentation

Author

Peng, Zelin, Xu, Zhengqin, Zeng, Zhilin, Wang, Yaoming, and Shen, Wei

Subjects
Computer Science - Computer Vision and Pattern Recognition
Abstract

Open-vocabulary semantic segmentation seeks to label each pixel in an image with arbitrary text descriptions. Vision-language foundation models, especially CLIP, have recently emerged as powerful tools for acquiring open-vocabulary capabilities. However, fine-tuning CLIP to equip it with pixel-level prediction ability often suffers three issues: 1) high computational cost, 2) misalignment between the two inherent modalities of CLIP, and 3) degraded generalization ability on unseen categories. To address these issues, we propose H-CLIP a symmetrical parameter-efficient fine-tuning (PEFT) strategy conducted in hyperspherical space for both of the two CLIP modalities. Specifically, the PEFT strategy is achieved by a series of efficient block-diagonal learnable transformation matrices and a dual cross-relation communication module among all learnable matrices. Since the PEFT strategy is conducted symmetrically to the two CLIP modalities, the misalignment between them is mitigated. Furthermore, we apply an additional constraint to PEFT on the CLIP text encoder according to the hyperspherical energy principle, i.e., minimizing hyperspherical energy during fine-tuning preserves the intrinsic structure of the original parameter space, to prevent the destruction of the generalization ability offered by the CLIP text encoder. Extensive evaluations across various benchmarks show that H-CLIP achieves new SOTA open-vocabulary semantic segmentation results while only requiring updating approximately 4% of the total parameters of CLIP.

Published
2024

3. SLSM : An Efficient Strategy for Lazy Schema Migration on Shared-Nothing Databases

Author

Zeng, Zhilin, Li, Hui, Gao, Xiyue, Zhang, Hui, Zhang, Huiquan, and Cui, Jiangtao

Subjects
Computer Science - Databases
Abstract

By introducing intermediate states for metadata changes and ensuring that at most two versions of metadata exist in the cluster at the same time, shared-nothing databases are capable of making online, asynchronous schema changes. However, this method leads to delays in the deployment of new schemas since it requires waiting for massive data backfill. To shorten the service vacuum period before the new schema is available, this paper proposes a strategy named SLSM for zero-downtime schema migration on shared-nothing databases. Based on the lazy migration of stand-alone databases, SLSM keeps the old and new schemas with the same data distribution, reducing the node communication overhead of executing migration transactions for shared-nothing databases. Further, SLSM combines migration transactions with user transactions by extending the distributed execution plan to allow the data involved in migration transactions to directly serve user transactions, greatly reducing the waiting time of user transactions. Experiments demonstrate that our strategy can greatly reduce the latency of user transactions and improve the efficiency of data migration compared to existing schemes.

Published
2024

5. Parameter Efficient Fine-tuning via Cross Block Orchestration for Segment Anything Model

Author

Peng, Zelin, Xu, Zhengqin, Zeng, Zhilin, Xie, Lingxi, Tian, Qi, and Shen, Wei

Subjects
Computer Science - Computer Vision and Pattern Recognition
Abstract

Parameter-efficient fine-tuning (PEFT) is an effective methodology to unleash the potential of large foundation models in novel scenarios with limited training data. In the computer vision community, PEFT has shown effectiveness in image classification, but little research has studied its ability for image segmentation. Fine-tuning segmentation models usually require a heavier adjustment of parameters to align the proper projection directions in the parameter space for new scenarios. This raises a challenge to existing PEFT algorithms, as they often inject a limited number of individual parameters into each block, which prevents substantial adjustment of the projection direction of the parameter space due to the limitation of Hidden Markov Chain along blocks. In this paper, we equip PEFT with a cross-block orchestration mechanism to enable the adaptation of the Segment Anything Model (SAM) to various downstream scenarios. We introduce a novel inter-block communication module, which integrates a learnable relation matrix to facilitate communication among different coefficient sets of each PEFT block's parameter space. Moreover, we propose an intra-block enhancement module, which introduces a linear projection head whose weights are generated from a hyper-complex layer, further enhancing the impact of the adjustment of projection directions on the entire parameter space. Extensive experiments on diverse benchmarks demonstrate that our proposed approach consistently improves the segmentation performance significantly on novel scenarios with only around 1K additional parameters., Comment: Accepted by CVPR2024

Published
2023

6. SAM-PARSER: Fine-tuning SAM Efficiently by Parameter Space Reconstruction

Author

Peng, Zelin, Xu, Zhengqin, Zeng, Zhilin, Yang, Xiaokang, and Shen, Wei

Subjects
Computer Science - Computer Vision and Pattern Recognition
Abstract

Segment Anything Model (SAM) has received remarkable attention as it offers a powerful and versatile solution for object segmentation in images. However, fine-tuning SAM for downstream segmentation tasks under different scenarios remains a challenge, as the varied characteristics of different scenarios naturally requires diverse model parameter spaces. Most existing fine-tuning methods attempt to bridge the gaps among different scenarios by introducing a set of new parameters to modify SAM's original parameter space. Unlike these works, in this paper, we propose fine-tuning SAM efficiently by parameter space reconstruction (SAM-PARSER), which introduce nearly zero trainable parameters during fine-tuning. In SAM-PARSER, we assume that SAM's original parameter space is relatively complete, so that its bases are able to reconstruct the parameter space of a new scenario. We obtain the bases by matrix decomposition, and fine-tuning the coefficients to reconstruct the parameter space tailored to the new scenario by an optimal linear combination of the bases. Experimental results show that SAM-PARSER exhibits superior segmentation performance across various scenarios, while reducing the number of trainable parameters by $\approx 290$ times compared with current parameter-efficient fine-tuning methods., Comment: Accepted by AAAI2024

Published
2023

7. Missing as Masking: Arbitrary Cross-Modal Feature Reconstruction for Incomplete Multimodal Brain Tumor Segmentation

Author

Zeng, Zhilin, Peng, Zelin, Yang, Xiaokang, Shen, Wei, Goos, Gerhard, Series Editor, Hartmanis, Juris, Founding Editor, Bertino, Elisa, Editorial Board Member, Gao, Wen, Editorial Board Member, Steffen, Bernhard, Editorial Board Member, Yung, Moti, Editorial Board Member, Linguraru, Marius George, editor, Dou, Qi, editor, Feragen, Aasa, editor, Giannarou, Stamatia, editor, Glocker, Ben, editor, Lekadir, Karim, editor, and Schnabel, Julia A., editor

Published
2024
Full Text
View/download PDF

26. Evaluation of Multi-model Integrated Forecast Method for Precipitation in Rainy Seasons of Guangdong-Hong Kong-Macao Greater Bay Area.

Author

ZHU Pengcheng, WANG Donghai, ZENG Zhilin, SUN Lei, LI Zhili, and ZI Guirong

Abstract

This study evaluated the precipitation forecasts in the Guangdong-Hong Kong-Macao Greater Bay Area during the 2018 South China rainy seasons (April to September) using data from the THORPEX Interactive Grand Global Ensemble dataset. The dataset comprised deterministic forecasts, ensemble forecasts, and ground precipitation observations from models of five organizations: the European Centre for Medium-Range Weather Forecasts, the China Meteorological Administration, the Japan Meteorological Agency, the National Centers for Environmental Prediction of the United States, and the UK Meteorological Office. Three multi-model integrated forecast methods, namely the multi-model ensemble average (EMN), the bias-removed ensemble average (BREM), and the sliding training period superensemble method (R_SUP), were employed for the evaluation. The results showed that, in general, the root mean square error of precipitation forecasts in the first rainy season of the Guangdong-Hong Kong- Macao Greater Bay Area was higher than that in the second rainy season, with an average difference of 2 mm. The forecasting ability of the multi-mode integrated forecasting method showed a continuous and stable decline trend in the first rainy season as the forecast lead time increased. In contrast, in the second rainy season, it showed a stable decline in the short term (24~72 hours) and remained stable in the medium term (72~168 hours). EMN, which has a relatively simple mathematical principle, showed the best comprehensive performance in the precipitation forecast of the two rainy seasons. BREM and R_SUP achieved slightly lower spatial average scores, but they still demonstrated good forecasting skills in predicting precipitation areas. [ABSTRACT FROM AUTHOR]

Published
2024
Full Text
View/download PDF

27. Pavement Crack Detection and Identification Based on Improved YOLOv8

Author

Tao, Xingzhen, Zeng, Zhilin, Zeng, Jun, Zhao, Taikang, and Dai, Qi

Abstract

Aiming at the problem of poor real-time performance and low precision of traditional pavement crack detection, an improved YOLOv8 algorithm is proposed to realize the automatic pavement crack detection and identification. Firstly, the crack image is manually labeled by LabelImg labeling software, and then the model parameters are obtained by improving YOLOv8 network. The improved method includes introducing a convolution block attention module (CBAM) to enhance the feature extraction ability, and optimizing the weighted intersection-to-union ratio (Wiou) loss function. Through comprehensive evaluation indexes (F1-measure and mAP@50-95%), the performance of the original model and the improved model in pavement crack detection and recognition is compared. The results show that the improved model has improved the detection precision, F1 value and mAP@50-95% value, which are 2.91%, 3.29% and 1.7% respectively. These results show that the improved model has significant practical significance in realizing the automatic identification of pavement cracks.

Published
2024
Full Text
View/download PDF

28. MULTI-SCALE OBSERVATIONAL CHARACTERISTICS AND EVOLUTION OF THE 22 MAY 2020 EXTREME RAINFALL IN THE GUANGDONG-HONG KONG-MACAO GREATER BAY AREA.

Author

LI Zhuofan, WANG Donghai, and ZENG Zhilin

Abstract

In the early morning of May 22, 2020, an extremely heavy rainfall process occurred in the Guangdong-Hong Kong-Macao Greater Bay Area. The maximum hourly and three-hour rainfall broke the historical record during the "Dragon Boat Water" period in Guangdong province. This paper uses multi-source observation data and ERA5 reanalysis data to study the evolution process and mesoscale characteristics of the mesoscale convective system (MCS) causing the local extreme precipitation. The results are shown as follows. (1) The heavy precipitation process mainly concentrated in the early morning hours, with the characteristics of short duration, extreme local rainfall intensity, and large accumulated rainfall. Two precipitation peaks appeared in the three-hour precipitation process. (2) This process is a torrential rain process in the warm sector. A low-level southwest monsoon provides sufficient warm and humid water vapor, and a low vortex shear line provides good uplifting conditions. The process of meso-α-scale monsoon cloud clusters was quasi-stationary and grew by upscaling into MCC, triggering the local nighttime extreme precipitation in the Greater Bay Area. (3) Two stages of peak precipitation in the process is closely related to the propagation of meso-γ-scale convective cells in the life cycle and the change of the motion vector. Several meso-γ-scale convective cells form a multi-cell storm pattern, which is arranged in the northwest-southeast lateral direction. These cells are triggered to develop successively and have their own ground convergence lines. The first stage of peak precipitation is caused by the convective cell moving to the southeast under the guidance of the ambient wind. Subsequently, the convergence lines of different cells are connected, and the propagation of convection to the southwest is significantly accelerated, which changes the movement vector of the convective system, thus causing the peak precipitation in the second stage. (4) There are obvious differences in the structure of the meso-γ-scale convective cells that cause the two periods of peak precipitation. The strong echoes of the peak precipitation in the first stage have a higher extension height with supercell-like structural characteristics such as inclination, inflow gap, and echo overhang, and the echoes of the second stage of peak precipitation are characterized by convective, low mass cores, causing more extreme and localized short-duration heavy precipitation. [ABSTRACT FROM AUTHOR]

Published
2023
Full Text
View/download PDF

34. SimSwap++: Towards Faster and High-Quality Identity Swapping

Author

Chen, Xuanhong, Ni, Bingbing, Liu, Yutian, Liu, Naiyuan, Zeng, Zhilin, and Wang, Hang

Abstract

Face identity editing (FIE) shows great value in AI content creation. Low-resolution FIE approaches have achieved tremendous progress, but high-quality FIE struggles. Two major challenges hinder higher-resolution and higher-performance development of FIE: lack of high-resolution dataset and unacceptable complexity forbidding for mobile platforms. To address both issues, we establish a novel large-scale, high-quality dataset tailored for FIE. Based on our SimSwap (Chen et al. 2020), we propose an upgraded version named SimSwap++ with significantly boosted model efficiency. SimSwap++ features two major innovations for high-performance model compression. First, a novel computational primitive named Conditional Dynamic Convolution (CD-Conv) is proposed to address the inefficiency of conditional schemes (e.g., AdaIN) in tiny models. CD-Conv achieves anisotropic processing and injection with significantly lower complexity compared to standard conditional operators, e.g., modulated convolution. Second, a Morphable Knowledge Distillation (MKD) is presented to further trim the overall model. Unlike conventional homogeneous teacher-student structures, MKD is designed to be heterogeneous and mutually compensable, endowing the student with the multi-path morphable property; thus, our student maximally inherits the teacher's knowledge after distillation while further reducing its complexity through structure re-parameterization. Extensive experiments demonstrate that our SimSwap++ achieves state-of-the-art performance (97.55% ID accuracy on FaceForensics++) with extremely low complexity (2.5 GFLOPs).

Published
2024
Full Text
View/download PDF

35. Increased Methyl-CpG-Binding Domain Protein 2 Promotes Cigarette Smoke-Induced Pulmonary Hypertension

Author

Wu, Jixing, Huang, Qian, Li, Qinghai, Gu, Yiya, Zhan, Yuan, Wang, Ting, Chen, Jinkun, Zeng, Zhilin, Lv, Yongman, Zhao, Jianping, Xia, Jie, and Xie, Jungang

Subjects
Cancer Research, Oncology
Abstract

Pulmonary hypertension (PH) is a chronic vascular proliferative disorder. While cigarette smoke (CS) plays a vital part in PH related to chronic obstructive pulmonary disease (COPD). Methyl-CpG-Binding Domain Protein 2 (MBD2) has been linked to multiple proliferative diseases. However, the specific mechanisms of MBD2 in CS-induced PH remain to be elucidated. Herein, the differential expression of MBD2 was tested between the controls and the PH patients’ pulmonary arteries, CS-exposed rat models’ pulmonary arteries, and primary human pulmonary artery smooth muscle cells (HPASMCs) following cigarette smoke extract (CSE) stimulation. As a result, PH patients and CS-induced rats and HPASMCs showed an increase in MBD2 protein expression compared with the controls. Then, MBD2 silencing was used to investigate the function of MBD2 on CSE-induced HPASMCs’ proliferation, migration, and cell cycle progression. As a consequence, CSE could induce HPASMCs’ increased proliferation and migration, and cell cycle transition, which were suppressed by MBD2 interference. Furthermore, RNA-seq, ChIP-qPCR, and MassARRAY were conducted to find out the downstream mechanisms of MBD2 for CS-induced pulmonary vascular remodeling. Subsequently, RNA-seq revealed MBD2 might affect the transcription of BMP2 gene, which furtherly altered the expression of BMP2 protein. ChIP-qPCR demonstrated MBD2 could bind BMP2’s promotor. MassARRAY indicated that MBD2 itself could not directly affect DNA methylation. In sum, our results indicate that increased MBD2 expression promotes CS-induced pulmonary vascular remodeling. The fundamental mechanisms may be that MBD2 can bind BMP2’s promoter and downregulate its expression. Thus, MBD2 may promote the occurrence of the CS-induced PH.

Published
2022
Full Text
View/download PDF

48. Twelve-Month Systemic Consequences of Coronavirus Disease 2019 (COVID-19) in Patients Discharged From Hospital: A Prospective Cohort Study in Wuhan, China

Author

Liu, Tingting, primary, Wu, Di, additional, Yan, Weiming, additional, Wang, Xiaojing, additional, Zhang, Xiaoyun, additional, Ma, Ke, additional, Chen, Huilong, additional, Zeng, Zhilin, additional, Qin, Yuanyuan, additional, Wang, Hongwu, additional, Xing, Mingyou, additional, Xu, Dong, additional, Li, Weina, additional, Ni, Ming, additional, Zhu, Lin, additional, Chen, Liang, additional, Chen, Guang, additional, Qi, Weipeng, additional, Wu, Ting, additional, Yu, Haijing, additional, Huang, Jiaquan, additional, Han, Meifang, additional, Zhu, Wenzhen, additional, Guo, Wei, additional, Luo, Xiaoping, additional, Chen, Tao, additional, and Ning, Qin, additional

Published
2021
Full Text
View/download PDF

50. Twelve-Month Systemic Consequences of Coronavirus Disease 2019 (COVID-19) in Patients Discharged From Hospital: A Prospective Cohort Study in Wuhan, China.

Author

Liu, Tingting, Wu, Di, Yan, Weiming, Wang, Xiaojing, Zhang, Xiaoyun, Ma, Ke, Chen, Huilong, Zeng, Zhilin, Qin, Yuanyuan, Wang, Hongwu, Xing, Mingyou, Xu, Dong, Li, Weina, Ni, Ming, Zhu, Lin, Chen, Liang, Chen, Guang, Qi, Weipeng, Wu, Ting, and Yu, Haijing

Subjects
HOSPITALS, COVID-19, SYMPTOMS, DISCHARGE planning, LONGITUDINAL method
Abstract

Background Follow-up study of coronavirus disease 2019 (COVID-19) survivors has rarely been reported. We aimed to investigate longitudinal changes in the characteristics of COVID-19 survivors after discharge. Methods A total of 594 COVID-19 survivors discharged from Tongji Hospital in Wuhan from February 10 to April 30, 2020 were included and followed up until May 17, 2021. Laboratory and radiological findings, pulmonary function tests, electrocardiogram, symptoms and signs were analyzed. Results 257 (51.2%) patients had at least one symptom at 3 months post-discharge, which decreased to 169 (40.0%) and 138 (28.4%) at 6-month and 12-month visit respectively. During follow-up period, insomnia, chest tightness, and fatigue were the most prevalent symptoms. Most laboratory parameters returned to normal, whereas increased incidence of abnormal liver and renal function and cardiovascular injury was evidenced after discharge. Fibrous stripes (213; 42.4%), pleural thickening and adhesions (188; 37.5%) and enlarged lymph nodes (120; 23.9%) were the most common radiographical findings at 3 months post-discharge. The abnormalities of pulmonary function included obstructive, restrictive, and mixed, which were 5.5%, 4.0%, 0.9% at 6 months post, and 1.9%, 4.7%, 0.2% at 12 months. Electrocardiogram abnormalities occurred in 256 (51.0%) patients at 3 months post-discharge, including arrhythmia, ST-T change and conduction block, which increased to 258 (61.1%) cases at 6-month visit and were maintained at high frequency (242;49.8%) at 12-month visit. Conclusions Physiological, laboratory, radiological, or electrocardiogram abnormalities, particularly those related to renal, cardiovascular, and liver functions are common in patients who recovered from coronavirus disease 2019 (COVID-19) up to 12 months post-discharge. [ABSTRACT FROM AUTHOR]

Published
2022
Full Text
View/download PDF

Catalog

Books, media, physical & digital resources
See catalog results