Geuntaek Lim
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Geuntaek Lim has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Yukyung Choi
Sejong University
6 shared publications - Joonsoo Kim
Electronics and Telecommunications Research Institute
3 shared publications - Gwangjin Lee
Department of Intelligent Mechatronics Engineering, Sejong University, Seoul 05006, Republic of Korea
2 shared publications - Joungil Yun
Electronics and Telecommunications Research Institute (ETRI), Yuseong-gu, Daejeon, Republic of Korea
2 shared publications - Won Jo
Department of Artificial Intelligence, Sejong University, Gwangjin-gu, Seoul, Republic of Korea
2 shared publications - Byungsoo Ko
School of Computing KAIST
1 shared publication
Latest Publications
- 9 September 2026
- Published by Springer Nature
- p. 608-626
Video-Oasis: Rethinking Evaluation of Video Understanding
- 9 September 2026
- Published by Springer Nature
- p. 608-626
Video-Oasis: Rethinking Evaluation of Video Understanding
- Open Access
- 27 April 2026
- Published by Elsevier in SSRN Electronic Journal
HypSGG: Open-Vocabulary Scene Graph Generation via Hyperbolic Representation and Hierarchical Supervision
Recent advances in vision-language models (VLMs), have expanded the scope of scene graph generation by enabling open-world visual relation detection. Nevertheless, VLMs still suffer from inherent limitations in spatial reasoning, particularly in modeling spatial configurations and attributive cues. Prior studies on open-vocabulary scene graph generation (OV-SGG) have tended to rely on adjustments of the VLM backbone, while leaving the fundamental challenges of the task largely unaddressed. We argue that the core bottleneck of OV-SGG lies in the combinatorial explosion of possible \textless subject-predicate-object\textgreater \, triplets, combined with the need for adaptability to novel environments. To overcome this issue, we introduce a hyperbolic geometry based framework, exploiting its exponential expansion property as a natural remedy for the combinatorial complexity. Furthermore, we propose a visual–semantic hierarchical supervision scheme that explicitly aligns visual features with textual semantics across multiple levels of granularity, thereby enhancing relational reasoning. Extensive experiments on widely used SGG benchmarks, evaluated in both closed- and open-vocabulary settings, demonstrate consistent improvements in accuracy and generalization over state-of-the-art methods.
- 27 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessHypSGG: Open-Vocabulary Scene Graph Generation via Hyperbolic Representation and Hierarchical Supervision
Recent advances in vision-language models (VLMs), have expanded the scope of scene graph generation by enabling open-world visual relation detection. Nevertheless, VLMs still suffer from inherent limitations in spatial reasoning, particularly in modeling spatial configurations and attributive cues. Prior studies on open-vocabulary scene graph generation (OV-SGG) have tended to rely on adjustments of the VLM backbone, while leaving the fundamental challenges of the task largely unaddressed. We argue that the core bottleneck of OV-SGG lies in the combinatorial explosion of possible \textless subject-predicate-object\textgreater \, triplets, combined with the need for adaptability to novel environments. To overcome this issue, we introduce a hyperbolic geometry based framework, exploiting its exponential expansion property as a natural remedy for the combinatorial complexity. Furthermore, we propose a visual–semantic hierarchical supervision scheme that explicitly aligns visual features with textual semantics across multiple levels of granularity, thereby enhancing relational reasoning. Extensive experiments on widely used SGG benchmarks, evaluated in both closed- and open-vocabulary settings, demonstrate consistent improvements in accuracy and generalization over state-of-the-art methods.
- Open Access
- 28 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 5507-5516
Probabilistic Vision-Language Representation for Weakly Supervised Temporal Action Localization
Weakly supervised temporal action localization (WTAL) aims to detect action instances in untrimmed videos using only video-level annotations. Since many existing works optimize WTAL models based on action classification labels, they encounter the task discrepancy problem (i.e., localization-by-classification). To tackle this issue, recent studies have attempted to utilize action category names as auxiliary semantic knowledge through vision-language pre-training (VLP). However, there are still areas where existing research falls short. Previous approaches primarily focused on leveraging textual information from language models but overlooked the alignment of dynamic human action and VLP knowledge in a joint space. Furthermore, the deterministic representation employed in previous studies struggles to capture fine-grained human motions. To address these problems, we propose a novel framework that aligns human action knowledge and VLP knowledge in a probabilistic embedding space. Moreover, we propose intra- and inter-distribution contrastive learning to enhance the probabilistic embedding space based on statistical similarities. Extensive experiments and ablation studies reveal that our method significantly outperforms all previous state-of-the-art methods. Code is available at https://github.com/sejong-rcv/PVLR.
- 28 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 5507-5516
Open AccessProbabilistic Vision-Language Representation for Weakly Supervised Temporal Action Localization
Weakly supervised temporal action localization (WTAL) aims to detect action instances in untrimmed videos using only video-level annotations. Since many existing works optimize WTAL models based on action classification labels, they encounter the task discrepancy problem (i.e., localization-by-classification). To tackle this issue, recent studies have attempted to utilize action category names as auxiliary semantic knowledge through vision-language pre-training (VLP). However, there are still areas where existing research falls short. Previous approaches primarily focused on leveraging textual information from language models but overlooked the alignment of dynamic human action and VLP knowledge in a joint space. Furthermore, the deterministic representation employed in previous studies struggles to capture fine-grained human motions. To address these problems, we propose a novel framework that aligns human action knowledge and VLP knowledge in a probabilistic embedding space. Moreover, we propose intra- and inter-distribution contrastive learning to enhance the probabilistic embedding space based on statistical similarities. Extensive experiments and ablation studies reveal that our method significantly outperforms all previous state-of-the-art methods. Code is available at https://github.com/sejong-rcv/PVLR.
- Open Access
- 24 March 2024
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 38 (3) , 2679-2687
VVS: Video-to-Video Retrieval with Irrelevant Frame Suppression
In content-based video retrieval (CBVR), dealing with large-scale collections, efficiency is as important as accuracy; thus, several video-level feature-based studies have actively been conducted. Nevertheless, owing to the severe difficulty of embedding a lengthy and untrimmed video into a single feature, these studies have been insufficient for accurate retrieval compared to frame-level feature-based studies. In this paper, we show that appropriate suppression of irrelevant frames can provide insight into the current obstacles of the video-level approaches. Furthermore, we propose a Video-to-Video Suppression network (VVS) as a solution. VVS is an end-to-end framework that consists of an easy distractor elimination stage to identify which frames to remove and a suppression weight generation stage to determine the extent to suppress the remaining frames. This structure is intended to effectively describe an untrimmed video with varying content and meaningless information. Its efficacy is proved via extensive experiments, and we show that our approach is not only state-of-the-art in video-level approaches but also has a fast inference time despite possessing retrieval capabilities close to those of frame-level approaches. Code is available at https://github.com/sejong-rcv/VVS
- 24 March 2024
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 38 (3) , 2679-2687
Open AccessVVS: Video-to-Video Retrieval with Irrelevant Frame Suppression
In content-based video retrieval (CBVR), dealing with large-scale collections, efficiency is as important as accuracy; thus, several video-level feature-based studies have actively been conducted. Nevertheless, owing to the severe difficulty of embedding a lengthy and untrimmed video into a single feature, these studies have been insufficient for accurate retrieval compared to frame-level feature-based studies. In this paper, we show that appropriate suppression of irrelevant frames can provide insight into the current obstacles of the video-level approaches. Furthermore, we propose a Video-to-Video Suppression network (VVS) as a solution. VVS is an end-to-end framework that consists of an easy distractor elimination stage to identify which frames to remove and a suppression weight generation stage to determine the extent to suppress the remaining frames. This structure is intended to effectively describe an untrimmed video with varying content and meaningless information. Its efficacy is proved via extensive experiments, and we show that our approach is not only state-of-the-art in video-level approaches but also has a fast inference time despite possessing retrieval capabilities close to those of frame-level approaches. Code is available at https://github.com/sejong-rcv/VVS
- Open Access
- 20 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 28466-28478
Simultaneous Video Retrieval and Alignment
With the growth of the video streaming industry, video retrieval and video alignment are facing high levels of demand. Several studies have demonstrated the feasibility of these methods for various problems related to video retrieval and alignment independently, but test in a unified framework has never been done. However, in real-world applications, it is also concurrently necessary not only to find which video pairs are similar (video retrieval) but also to align the positions of the pair that are related (video alignment). In this paper, we present a new task that simultaneously retrieves and aligns videos. As a solution to this task, a simultaneous retrieval and alignment framework abbreviated as SRA is proposed, which is a two-stage approach consisting of a foreground proposal stage and a downstream stage to efficiently process untrimmed videos. Furthermore, two criteria are suggested to support the new task: a metric mAP@J assessing how highly the related videos are ranked and how well relevant positions are assigned in that video, and a dataset FIVR+A that includes video-level relationships and hierarchical segment-level annotations. Finally, we conduct multi-pronged analyses to assess how our approach handles the new task in various experiments.
- 20 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 28466-28478
Open AccessSimultaneous Video Retrieval and Alignment
With the growth of the video streaming industry, video retrieval and video alignment are facing high levels of demand. Several studies have demonstrated the feasibility of these methods for various problems related to video retrieval and alignment independently, but test in a unified framework has never been done. However, in real-world applications, it is also concurrently necessary not only to find which video pairs are similar (video retrieval) but also to align the positions of the pair that are related (video alignment). In this paper, we present a new task that simultaneously retrieves and aligns videos. As a solution to this task, a simultaneous retrieval and alignment framework abbreviated as SRA is proposed, which is a two-stage approach consisting of a foreground proposal stage and a downstream stage to efficiently process untrimmed videos. Furthermore, two criteria are suggested to support the new task: a metric mAP@J assessing how highly the related videos are ranked and how well relevant positions are assigned in that video, and a dataset FIVR+A that includes video-level relationships and hierarchical segment-level annotations. Finally, we conduct multi-pronged analyses to assess how our approach handles the new task in various experiments.
- Open Access
- 11 April 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 38973-38981
Exploring the Temporal Cues to Enhance Video Retrieval on Standardized CDVA
As the demand for large-scale video analysis increases, video retrieval research is also becoming more active. In 2014, ISO/IEC MPEG began standardizing compact descriptors for video analysis, known as CDVA, and it is now adopted as a standard. However, the standardized CDVA is not easily compared to other methods because the MPEG-CDVA dataset used for performance verification is not disclosed, despite the fact that follow-up studies are underway with multiple versions of the CDVA experimental model. In addition, analyses of modules constituting the CDVA framework are insufficient in previous studies. Therefore, we conduct self-evaluations of CDVA to analyze the impact of each module on the retrieval task. Furthermore, to overcome the obstacles identified through these self-evaluations, we suggest temporal nested invariance pooling, abbreviated as TNIP, which implies temporal robustness realized by improving nested invariance pooling, abbreviated as NIP, one of the features in CDVA. Finally, benchmarks of the existing CDVA and the proposed approach are provided on several public datasets. Through this, we show that the CDVA framework is capable of boosting the retrieval performance if utilizing the proposed approach.
- 11 April 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 38973-38981
Open AccessExploring the Temporal Cues to Enhance Video Retrieval on Standardized CDVA
As the demand for large-scale video analysis increases, video retrieval research is also becoming more active. In 2014, ISO/IEC MPEG began standardizing compact descriptors for video analysis, known as CDVA, and it is now adopted as a standard. However, the standardized CDVA is not easily compared to other methods because the MPEG-CDVA dataset used for performance verification is not disclosed, despite the fact that follow-up studies are underway with multiple versions of the CDVA experimental model. In addition, analyses of modules constituting the CDVA framework are insufficient in previous studies. Therefore, we conduct self-evaluations of CDVA to analyze the impact of each module on the retrieval task. Furthermore, to overcome the obstacles identified through these self-evaluations, we suggest temporal nested invariance pooling, abbreviated as TNIP, which implies temporal robustness realized by improving nested invariance pooling, abbreviated as NIP, one of the features in CDVA. Finally, benchmarks of the existing CDVA and the proposed approach are provided on several public datasets. Through this, we show that the CDVA framework is capable of boosting the retrieval performance if utilizing the proposed approach.