Sungjei Kim
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Sungjei Kim has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
- Jinwoo Jeong
Intelligent Semiconductor Device Research Center, Korea Electronics Technology Institute, Seongnam-si, Republic of Korea
4 shared publications - Hyeon-Cheol Moon
Intelligent Semiconductor Device Research Center, Korea Electronics Technology Institute, Seongnam-si, Republic of Korea
2 shared publications Jinwoo Jeong
2 shared publications- Kihwan Yoon
University of Seoul
2 shared publications - Yong Han Kim
University of Seoul
2 shared publications - Chae-Yeong Song
School of Computer Science and Engineering, Kyungpook National University, Daegu, South Korea
1 shared publication
Latest Publications
- 6 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28 (15209210) , 3135-3146
Compression Framework for Light 3D Scene Graph Generation via Pruning-as-Search and Distillation
3D scene graph generation (3DSGG), which involves classifying objects and predicates, is an emerging topic in 3D scene understanding. Recent studies leveraging graph neural networks (GNNs) have introduced sophisticated architectures that enhance classification performance. However, since GNNs serve as the core and constitute the majority of parameters in 3DSGG models, their computational demands substantially increase overall complexity, which makes it difficult to determine the optimal model capacity. In this paper, we propose the first compression framework for lightweight 3DSGG models, based on pruning-as-search and knowledge distillation. This framework integrates multiple strategies and modules. In phase 1, the framework identifies the optimal compression ratio through pruning-as-search. In phase 2, to mitigate the accuracy loss incurred during compression, we employ structured pruning and a novel knowledge distillation strategy that effectively transfers precise information from the teacher to the compressed model. Experimental results show that our approach reduces model size by more than half while improving classification accuracy. Code is available at https://github.com/hojunking/3DSGG-compression.
- 6 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28 (15209210) , 3135-3146
Compression Framework for Light 3D Scene Graph Generation via Pruning-as-Search and Distillation
3D scene graph generation (3DSGG), which involves classifying objects and predicates, is an emerging topic in 3D scene understanding. Recent studies leveraging graph neural networks (GNNs) have introduced sophisticated architectures that enhance classification performance. However, since GNNs serve as the core and constitute the majority of parameters in 3DSGG models, their computational demands substantially increase overall complexity, which makes it difficult to determine the optimal model capacity. In this paper, we propose the first compression framework for lightweight 3DSGG models, based on pruning-as-search and knowledge distillation. This framework integrates multiple strategies and modules. In phase 1, the framework identifies the optimal compression ratio through pruning-as-search. In phase 2, to mitigate the accuracy loss incurred during compression, we employ structured pruning and a novel knowledge distillation strategy that effectively transfers precise information from the teacher to the compressed model. Experimental results show that our approach reduces model size by more than half while improving classification accuracy. Code is available at https://github.com/hojunking/3DSGG-compression.
- 22 November 2024
- Published by Springer Nature
- p. 461-477
IAM-VFI : Interpolate Any Motion for Video Frame Interpolation with Motion Complexity Map
- 22 November 2024
- Published by Springer Nature
- p. 461-477
IAM-VFI : Interpolate Any Motion for Video Frame Interpolation with Motion Complexity Map
- Open Access
- 30 September 2024
- Published by The Korean Institute of Broadcast and Media Engineers in Journal of Broadcast Engineering
- Vol. 29 (5) , 729-738
Proposal of a Recommendation System for Time-space Matching of Motion Capture Device and Camera
- 30 September 2024
- Published by The Korean Institute of Broadcast and Media Engineers in Journal of Broadcast Engineering
- Vol. 29 (5) , 729-738
Open AccessProposal of a Recommendation System for Time-space Matching of Motion Capture Device and Camera
- Open Access
- 30 November 2023
- Published by The Korean Institute of Broadcast and Media Engineers in Journal of Broadcast Engineering
- Vol. 28 (6) , 709-719
Contrastive Learning based Self-Distillation for Lightweight Single Image Super-Resolution Network
- 30 November 2023
- Published by The Korean Institute of Broadcast and Media Engineers in Journal of Broadcast Engineering
- Vol. 28 (6) , 709-719
Open AccessContrastive Learning based Self-Distillation for Lightweight Single Image Super-Resolution Network
- Open Access
- 23 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 131885-131896
Feature-Domain Adaptive Contrastive Distillation for Efficient Single Image Super-Resolution
Convolutional neural network-based single image super-resolution (SISR) involves numerous parameters and high computational expenses to ensure improved performance, limiting its applicability in resource-constrained devices such as mobile phones. Knowledge distillation (KD), which transfers useful knowledge from a teacher network to a student network, has been investigated as a method to make networks more efficient in terms of performance. To this end, feature distillation (FD) has been utilized in KD to minimize the Euclidean distance-based loss of feature maps between teacher and student networks. However, this technique does not adequately consider the effective and meaningful delivery of knowledge from the teacher to the student network to improve the latter’s performance under given network capacity constraints. In this study, we propose a feature-domain adaptive contrastive distillation (FACD) method to train lightweight student SISR networks efficiently. We highlight the limitations of existing FD methods in terms of Euclidean distance-based loss, and propose a feature-domain contrastive loss, which causes student networks to learn richer information from the teacher’s representation in the feature domain. We also implement adaptive distillation that performs distillation selectively depending on the conditions of the training patches. Experimental results showed that the proposed FACD scheme improves student enhanced deep residual networks and residual channel attention networks not only in terms of the peak signal-to-noise ratio on all benchmark datasets and scales but also in terms of subjective image quality, compared to the conventional FD approaches. Code will be release at https://github.com/hcmoon0613/FACD.
- 23 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 131885-131896
Open AccessFeature-Domain Adaptive Contrastive Distillation for Efficient Single Image Super-Resolution
Convolutional neural network-based single image super-resolution (SISR) involves numerous parameters and high computational expenses to ensure improved performance, limiting its applicability in resource-constrained devices such as mobile phones. Knowledge distillation (KD), which transfers useful knowledge from a teacher network to a student network, has been investigated as a method to make networks more efficient in terms of performance. To this end, feature distillation (FD) has been utilized in KD to minimize the Euclidean distance-based loss of feature maps between teacher and student networks. However, this technique does not adequately consider the effective and meaningful delivery of knowledge from the teacher to the student network to improve the latter’s performance under given network capacity constraints. In this study, we propose a feature-domain adaptive contrastive distillation (FACD) method to train lightweight student SISR networks efficiently. We highlight the limitations of existing FD methods in terms of Euclidean distance-based loss, and propose a feature-domain contrastive loss, which causes student networks to learn richer information from the teacher’s representation in the feature domain. We also implement adaptive distillation that performs distillation selectively depending on the conditions of the training patches. Experimental results showed that the proposed FACD scheme improves student enhanced deep residual networks and residual channel attention networks not only in terms of the peak signal-to-noise ratio on all benchmark datasets and scales but also in terms of subjective image quality, compared to the conventional FD approaches. Code will be release at https://github.com/hcmoon0613/FACD.
- Open Access
- 13 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 71994-72006
Textural Detail Preservation Network for Video Frame Interpolation
The subjective image quality of the Video Frame Interpolation (VFI) result depends on whether image features such as edges, textures and blobs are preserved. With the development of deep learning, various algorithms have been proposed and the objective results of VFI have significantly improved. Moreover, perceptual loss has been used in a method that enhances subjective quality by preserving the features of the image, and as a result, the subjective quality is improved. Despite the quality enhancements achieved in VFI, no analysis has been performed to preserve specific features in the interpolated frames. Therefore, we conducted an analysis to preserve textural detail, such as film grain noise, which can represent the texture of an image, and weak textures, such as droplets or particles. Based on our analysis, we identify the importance of synthesis networks in textural detail preservation and propose an enhanced synthesis network, the Textural Detail Preservation Network (TDPNet). Furthermore, based on our analysis, we propose a Perceptual Training Method (PTM) to address the issue of degraded Peak Signal-to-Noise Ratio (PSNR) when simply applying perceptual loss and to preserve more textural detail. We also propose a Multi-scale Resolution Training Method (MRTM) to address the issue of poor performance when testing datasets with a resolution different from that of the training dataset. The experimental results of the proposed network was outperformed in LPIPS and DISTS on the Vimeo90K, HD, SNU-FILM and UVG datasets compared with the state-of-the-art VFI algorithms, and the subjective results were also outperformed. Furthermore, applying PTM improved PSNR results by an average of 0.293dB compared to simply applying perceptual loss.
- 13 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 71994-72006
Open AccessTextural Detail Preservation Network for Video Frame Interpolation
The subjective image quality of the Video Frame Interpolation (VFI) result depends on whether image features such as edges, textures and blobs are preserved. With the development of deep learning, various algorithms have been proposed and the objective results of VFI have significantly improved. Moreover, perceptual loss has been used in a method that enhances subjective quality by preserving the features of the image, and as a result, the subjective quality is improved. Despite the quality enhancements achieved in VFI, no analysis has been performed to preserve specific features in the interpolated frames. Therefore, we conducted an analysis to preserve textural detail, such as film grain noise, which can represent the texture of an image, and weak textures, such as droplets or particles. Based on our analysis, we identify the importance of synthesis networks in textural detail preservation and propose an enhanced synthesis network, the Textural Detail Preservation Network (TDPNet). Furthermore, based on our analysis, we propose a Perceptual Training Method (PTM) to address the issue of degraded Peak Signal-to-Noise Ratio (PSNR) when simply applying perceptual loss and to preserve more textural detail. We also propose a Multi-scale Resolution Training Method (MRTM) to address the issue of poor performance when testing datasets with a resolution different from that of the training dataset. The experimental results of the proposed network was outperformed in LPIPS and DISTS on the Vimeo90K, HD, SNU-FILM and UVG datasets compared with the state-of-the-art VFI algorithms, and the subjective results were also outperformed. Furthermore, applying PTM improved PSNR results by an average of 0.293dB compared to simply applying perceptual loss.
- Open Access
Feature-domain Adaptive Contrastive Distillation for Efficient Single Image Super-Resolution
Recently, CNN-based SISR has numerous parameters and high computational cost to achieve better performance, limiting its applicability to resource-constrained devices such as mobile. As one of the methods to make the network efficient, Knowledge Distillation (KD), which transfers teacher's useful knowledge to student, is currently being studied. More recently, KD for SISR utilizes Feature Distillation (FD) to minimize the Euclidean distance loss of feature maps between teacher and student networks, but it does not sufficiently consider how to effectively and meaningfully deliver knowledge from teacher to improve the student performance at given network capacity constraints. In this paper, we propose a feature-domain adaptive contrastive distillation (FACD) method for efficiently training lightweight student SISR networks. We show the limitations of the existing FD methods using Euclidean distance loss, and propose a feature-domain contrastive loss that makes a student network learn richer information from the teacher's representation in the feature domain. In addition, we propose an adaptive distillation that selectively applies distillation depending on the conditions of the training patches. The experimental results show that the student EDSR and RCAN networks with the proposed FACD scheme improves not only the PSNR performance of the entire benchmark datasets and scales, but also the subjective image quality compared to the conventional FD approaches.
Open AccessFeature-domain Adaptive Contrastive Distillation for Efficient Single Image Super-Resolution
Recently, CNN-based SISR has numerous parameters and high computational cost to achieve better performance, limiting its applicability to resource-constrained devices such as mobile. As one of the methods to make the network efficient, Knowledge Distillation (KD), which transfers teacher's useful knowledge to student, is currently being studied. More recently, KD for SISR utilizes Feature Distillation (FD) to minimize the Euclidean distance loss of feature maps between teacher and student networks, but it does not sufficiently consider how to effectively and meaningfully deliver knowledge from teacher to improve the student performance at given network capacity constraints. In this paper, we propose a feature-domain adaptive contrastive distillation (FACD) method for efficiently training lightweight student SISR networks. We show the limitations of the existing FD methods using Euclidean distance loss, and propose a feature-domain contrastive loss that makes a student network learn richer information from the teacher's representation in the feature domain. In addition, we propose an adaptive distillation that selectively applies distillation depending on the conditions of the training patches. The experimental results show that the student EDSR and RCAN networks with the proposed FACD scheme improves not only the PSNR performance of the entire benchmark datasets and scales, but also the subjective image quality compared to the conventional FD approaches.
- Open Access
- 26 December 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 4972-4982
Skip-Concatenated Image Super-Resolution Network for Mobile Devices
Single-image super-resolution technology has been widely studied in various applications to improve the quality and resolution of degraded images acquired from noise-sensitive low-resolution sensors. As most studies on single-image super-resolution focused on the development of deep learning networks operating on high-performance GPUs, this study proposed an efficient and lightweight super-resolution network that enables real-time performance on mobile devices. To replace the relatively slow element-wise addition layer on mobile devices, we introduced a skip connection layer by directly concatenating a low-resolution input image with an intermediate feature map. In addition, we introduced weighted clipping to reduce the quantization errors commonly encountered during float-to-int8 model conversion. Moreover, a reparameterization method was selectively applied without increasing the cost in terms of inference time and number of parameters. Based on the contributions, the proposed network has been recognized as the best solution in Mobile AI & AIM 2022 Real-Time Single-Image Super-Resolution Challenge with PSNR of 30.03 dB and NPU runtime of 19.20 ms.
- 26 December 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 4972-4982
Open AccessSkip-Concatenated Image Super-Resolution Network for Mobile Devices
Single-image super-resolution technology has been widely studied in various applications to improve the quality and resolution of degraded images acquired from noise-sensitive low-resolution sensors. As most studies on single-image super-resolution focused on the development of deep learning networks operating on high-performance GPUs, this study proposed an efficient and lightweight super-resolution network that enables real-time performance on mobile devices. To replace the relatively slow element-wise addition layer on mobile devices, we introduced a skip connection layer by directly concatenating a low-resolution input image with an intermediate feature map. In addition, we introduced weighted clipping to reduce the quantization errors commonly encountered during float-to-int8 model conversion. Moreover, a reparameterization method was selectively applied without increasing the cost in terms of inference time and number of parameters. Based on the contributions, the proposed network has been recognized as the best solution in Mobile AI & AIM 2022 Real-Time Single-Image Super-Resolution Challenge with PSNR of 30.03 dB and NPU runtime of 19.20 ms.