Weiwei Kong
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Weiwei Kong has a total of 20 co-authors. Here are the top co-authors by number of shared publications:
Yang Lei
6 shared publications- Leilei He
School of Computer Science & Technology, Xi’an University of Posts & Telecommunications, Xian, China
2 shared publications - Yusheng Du
School of Computer Science & Technology, Xi’an University of Posts & Telecommunications, Xian, China
2 shared publications - Zejiang Li
School of Computer Science & Technology, Xi’an University of Posts & Telecommunications, Xian, China
2 shared publications Chi Li
School of Computer Science and Technology, Xi'an University of Posts and Telecommunications, Xi'an, China
1 shared publication- Chi Li
Xi’an University of Posts and Telecommunications
1 shared publication
Latest Publications
- 26 September 2026
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 130
DEAMNet: a dual-branch mamba-CNN encoder with wavelet boundary awareness for colorectal polyp segmentation
- 26 September 2026
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 130
DEAMNet: a dual-branch mamba-CNN encoder with wavelet boundary awareness for colorectal polyp segmentation
- 1 April 2026
- Published by Springer Nature in International Journal of Machine Learning and Cybernetics
- Vol. 17 (5) , 247
Medscale+: A medical image segmentation method based on multi-scale feature fusion and Monte Carlo Dropout
- 1 April 2026
- Published by Springer Nature in International Journal of Machine Learning and Cybernetics
- Vol. 17 (5) , 247
Medscale+: A medical image segmentation method based on multi-scale feature fusion and Monte Carlo Dropout
- Open Access
- 20 September 2024
- Published by Association for Computing Machinery (ACM)
- p. 523-528
Fitness Action Counting Algorithm based on Pose estimation
- 20 September 2024
- Published by Association for Computing Machinery (ACM)
- p. 523-528
Open AccessFitness Action Counting Algorithm based on Pose estimation
- Open Access
- 20 September 2024
- Published by Association for Computing Machinery (ACM)
- p. 426-432
A NeRF Rendering Method for Low-Light Scenes Using Occlusion Perception and Network Decomposition
- 20 September 2024
- Published by Association for Computing Machinery (ACM)
- p. 426-432
Open AccessA NeRF Rendering Method for Low-Light Scenes Using Occlusion Perception and Network Decomposition
- Open Access
- 24 July 2024
- Published by MDPI AG in Electronics
- Vol. 13 (15) , 2915
Improved 3D Object Detection Based on PointPillars
Despite the recent advancements in 3D object detection, the conventional 3D point cloud object detection algorithms have been found to exhibit limited accuracy for the detection of small objects. To address the challenge of poor detection of small-scale objects, this paper adopts the PointPillars algorithm as the baseline model and proposes a two-stage 3D target detection approach. As a cutting-edge solution, point cloud processing is performed using Transformer models. Additionally, a redefined attention mechanism is introduced to further enhance the detection capabilities of the algorithm. In the first stage, the algorithm uses PointPillars as the baseline model. The central concept of this algorithm is to transform the point cloud space into equal-sized columns. During the feature extraction stage, when the features from all cylinders are transformed into pseudo-images, the proposed algorithm incorporates attention mechanisms adapted from the Squeeze-and-Excitation (SE) method to emphasize and suppress feature information. Furthermore, the 2D convolution of the traditional backbone network is replaced by dynamic convolution. Concurrently, the addition of the attention mechanism further improves the feature representation ability of the network. In the second phase, the candidate frames generated in the first phase are refined using a Transformer-based approach. The proposed algorithm applies channel weighting in the decoder to enhance channel information, leading to improved detection accuracy and reduced false detections. The encoder constructs the initial point features from the candidate frames for encoding. Meanwhile, the decoder applies channel weighting to enhance the channel information, thereby improving the detection accuracy and reducing false detections. In the KITTI dataset, the experimental results verify the effectiveness of this method in small objects detection. Experimental results show that the proposed method significantly improves the detection capability of small objects compared with the baseline PointPillars. In concrete terms, in the moderate difficulty detection category, cars, pedestrians, and cyclists average precision (AP) values increased by 5.30%, 8.1%, and 10.6%, respectively. Moreover, the proposed method surpasses existing mainstream approaches in the cyclist category.
- 24 July 2024
- Published by MDPI AG in Electronics
- Vol. 13 (15) , 2915
Open AccessImproved 3D Object Detection Based on PointPillars
Despite the recent advancements in 3D object detection, the conventional 3D point cloud object detection algorithms have been found to exhibit limited accuracy for the detection of small objects. To address the challenge of poor detection of small-scale objects, this paper adopts the PointPillars algorithm as the baseline model and proposes a two-stage 3D target detection approach. As a cutting-edge solution, point cloud processing is performed using Transformer models. Additionally, a redefined attention mechanism is introduced to further enhance the detection capabilities of the algorithm. In the first stage, the algorithm uses PointPillars as the baseline model. The central concept of this algorithm is to transform the point cloud space into equal-sized columns. During the feature extraction stage, when the features from all cylinders are transformed into pseudo-images, the proposed algorithm incorporates attention mechanisms adapted from the Squeeze-and-Excitation (SE) method to emphasize and suppress feature information. Furthermore, the 2D convolution of the traditional backbone network is replaced by dynamic convolution. Concurrently, the addition of the attention mechanism further improves the feature representation ability of the network. In the second phase, the candidate frames generated in the first phase are refined using a Transformer-based approach. The proposed algorithm applies channel weighting in the decoder to enhance channel information, leading to improved detection accuracy and reduced false detections. The encoder constructs the initial point features from the candidate frames for encoding. Meanwhile, the decoder applies channel weighting to enhance the channel information, thereby improving the detection accuracy and reducing false detections. In the KITTI dataset, the experimental results verify the effectiveness of this method in small objects detection. Experimental results show that the proposed method significantly improves the detection capability of small objects compared with the baseline PointPillars. In concrete terms, in the moderate difficulty detection category, cars, pedestrians, and cyclists average precision (AP) values increased by 5.30%, 8.1%, and 10.6%, respectively. Moreover, the proposed method surpasses existing mainstream approaches in the cyclist category.
- Open Access
- 17 April 2024
- Published by Springer Nature in Research Square
A Two-stage Shadow Removal Algorithm Based on Recurrent Attention Network
The presence of shadows in an image obscures some information, hindering the subsequent image-processing task. So correct identification of shadow regions will greatly upgrade the performance of shadow removal. Obviously, the more accurate the identification of shadow regions, the better the shadow removal performance will be. Based on this, this paper designs a two-stage shadow removal algorithm (TS-RAN) based on the recurrent attention network, including the shadow detection stage and shadow removal stage. Firstly, a single shadow image is fed into the shadow detection stage, which generates a rough attention map under the joint action of the recurrent attention network and luminance prediction module. Secondly, the results generated in the previous stage are transferred to the shadow removal stage and the luminance estimation module for fine shadow removal. In addition, we designed a residual estimation module to remove possible artifacts caused by shadow residuals. Experiments show that our method achieves significant performance on the LRSS, ISTD, and WSRD datasets.
- 17 April 2024
- Published by Springer Nature in Research Square
Open AccessA Two-stage Shadow Removal Algorithm Based on Recurrent Attention Network
The presence of shadows in an image obscures some information, hindering the subsequent image-processing task. So correct identification of shadow regions will greatly upgrade the performance of shadow removal. Obviously, the more accurate the identification of shadow regions, the better the shadow removal performance will be. Based on this, this paper designs a two-stage shadow removal algorithm (TS-RAN) based on the recurrent attention network, including the shadow detection stage and shadow removal stage. Firstly, a single shadow image is fed into the shadow detection stage, which generates a rough attention map under the joint action of the recurrent attention network and luminance prediction module. Secondly, the results generated in the previous stage are transferred to the shadow removal stage and the luminance estimation module for fine shadow removal. In addition, we designed a residual estimation module to remove possible artifacts caused by shadow residuals. Experiments show that our method achieves significant performance on the LRSS, ISTD, and WSRD datasets.
- Open Access
- 22 September 2023
- Published by Association for Computing Machinery (ACM)
- p. 409-414
Zero-shot Image Caption Enhancement using Large-Scale Language Models
Large-scale language models (LLMs) have demonstrated impressive zero-shot learning generalization capabilities for new language tasks. However, applying LLMs to the image captioning problem still poses challenges. This is primarily due to the modality gap between image captioning and LLMs. Existing methods address the modality gap by training language and visual models end-to-end, but this approach incurs significant computational costs. To address this challenge, we propose a novel approach that combines multiple caption generation and fusion of image captions, providing cues for decoupling the modalities and tasks of image captioning from LLMs without the need for end-to-end training. Specifically, we generate multiple captions, relevant questions, and their answers for each image. Then, using prompt templates, we input this information into a large-scale pre-trained language model to generate captions. This approach offers several benefits: Flexibility: It allows for the flexible use of various LLMs to perform the image captioning task. Cost Efficiency: By avoiding end-to-end training, significantly reduces the computational costs of zero-shot learning with LLMs. Performance Improvement: Our proposed method outperforms previous approaches, demonstrating enhanced performance in generating image captions. By adopting this approach, we aim to bridge the gap between LLMs and the image captioning task, providing a cost-effective solution that achieves improved performance without the need for resource-intensive end-to-end training.
- 22 September 2023
- Published by Association for Computing Machinery (ACM)
- p. 409-414
Open AccessZero-shot Image Caption Enhancement using Large-Scale Language Models
Large-scale language models (LLMs) have demonstrated impressive zero-shot learning generalization capabilities for new language tasks. However, applying LLMs to the image captioning problem still poses challenges. This is primarily due to the modality gap between image captioning and LLMs. Existing methods address the modality gap by training language and visual models end-to-end, but this approach incurs significant computational costs. To address this challenge, we propose a novel approach that combines multiple caption generation and fusion of image captions, providing cues for decoupling the modalities and tasks of image captioning from LLMs without the need for end-to-end training. Specifically, we generate multiple captions, relevant questions, and their answers for each image. Then, using prompt templates, we input this information into a large-scale pre-trained language model to generate captions. This approach offers several benefits: Flexibility: It allows for the flexible use of various LLMs to perform the image captioning task. Cost Efficiency: By avoiding end-to-end training, significantly reduces the computational costs of zero-shot learning with LLMs. Performance Improvement: Our proposed method outperforms previous approaches, demonstrating enhanced performance in generating image captions. By adopting this approach, we aim to bridge the gap between LLMs and the image captioning task, providing a cost-effective solution that achieves improved performance without the need for resource-intensive end-to-end training.
- Open Access
- 22 September 2023
- Published by Association for Computing Machinery (ACM)
- p. 345-351
A deep bilateral learning based GAN for non-homogeneous dehazing
Affected by haze, images often face color distortion, resolution reduction and other image quality degradation problems. The existing dehazing methods based on convolutional neural network(CNN) often perform well on large-scale synthetic datasets, but lack robustness in the processing of real haze images. This is because haze images in reality are often non-homogeneous. Due to the fact that the haze texture faced by the real haze processing is more complex, it is easier to destroy the texture details. Meanwhile, the paired training images are difficult to collect, and the small-scale data set is easy to lead to overfitting. To address these challenges, we propose a dehazing approach based on ensemble learning, DB-GAN, which uses Res2Net pre-trained by ImageNet as the encoder in the knowledge adaptation branch to improve the generalization ability of the network and avoid overfitting. In the data fitting branch, deep bilateral learning is used to learn the structure of the features from the full-resolution and low-resolution inputs, respectively, to better learn the color features and boundary features. We then map the different features by a fusion tail. Finally, we demonstrate the effectiveness of our approach through extensive experimental results.
- 22 September 2023
- Published by Association for Computing Machinery (ACM)
- p. 345-351
Open AccessA deep bilateral learning based GAN for non-homogeneous dehazing
Affected by haze, images often face color distortion, resolution reduction and other image quality degradation problems. The existing dehazing methods based on convolutional neural network(CNN) often perform well on large-scale synthetic datasets, but lack robustness in the processing of real haze images. This is because haze images in reality are often non-homogeneous. Due to the fact that the haze texture faced by the real haze processing is more complex, it is easier to destroy the texture details. Meanwhile, the paired training images are difficult to collect, and the small-scale data set is easy to lead to overfitting. To address these challenges, we propose a dehazing approach based on ensemble learning, DB-GAN, which uses Res2Net pre-trained by ImageNet as the encoder in the knowledge adaptation branch to improve the generalization ability of the network and avoid overfitting. In the data fitting branch, deep bilateral learning is used to learn the structure of the features from the full-resolution and low-resolution inputs, respectively, to better learn the color features and boundary features. We then map the different features by a fusion tail. Finally, we demonstrate the effectiveness of our approach through extensive experimental results.
- Open Access
- 13 June 2023
- Published by MDPI AG in Electronics
- Vol. 12 (12) , 2659
Medical Image Fusion Using SKWGF and SWF in Framelet Transform Domain
Accurately localizing and describing patients’ lesions has long been considered a crucial aspect of clinical diagnosis. The fusion of multimodal medical images provides a feasible solution to the above problem. Unfortunately, the trade-off between the fusion performance and heavy computation overhead remains a challenge. In this paper, a novel and effective fusion method for multimodal medical images is proposed. Firstly, framelet transform (FT) is introduced to decompose the source images into a series of low and high frequency sub-images. Next, we utilize the benefits of both steering kernel weighted guided filtering and side window filtering to successfully fuse sub-images. Finally, the inverse FT is employed to reconstruct the final fused image. To verify the effectiveness of the proposed fusion method, we fused several pairs of medical images covering different modalities in simulation experiments. The experimental results demonstrate that the proposed method yields better performance than current representative ones in terms of both visual quality and quantitative evaluation.
- 13 June 2023
- Published by MDPI AG in Electronics
- Vol. 12 (12) , 2659
Open AccessMedical Image Fusion Using SKWGF and SWF in Framelet Transform Domain
Accurately localizing and describing patients’ lesions has long been considered a crucial aspect of clinical diagnosis. The fusion of multimodal medical images provides a feasible solution to the above problem. Unfortunately, the trade-off between the fusion performance and heavy computation overhead remains a challenge. In this paper, a novel and effective fusion method for multimodal medical images is proposed. Firstly, framelet transform (FT) is introduced to decompose the source images into a series of low and high frequency sub-images. Next, we utilize the benefits of both steering kernel weighted guided filtering and side window filtering to successfully fuse sub-images. Finally, the inverse FT is employed to reconstruct the final fused image. To verify the effectiveness of the proposed fusion method, we fused several pairs of medical images covering different modalities in simulation experiments. The experimental results demonstrate that the proposed method yields better performance than current representative ones in terms of both visual quality and quantitative evaluation.
- 30 January 2023
- Published by Springer Nature
- p. 656-662
Image Fusion Method Based on Improved Framelet Transform
- 30 January 2023
- Published by Springer Nature
- p. 656-662
Image Fusion Method Based on Improved Framelet Transform