Junyeong Kim
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
Research Interests
Junyeong Kim has a total of 72 co-authors. Here are the top co-authors by number of shared publications:
- Chang Dong Yoo
KAIST
11 shared publications - Sunjae Yoon
Chung-Ang University
11 shared publications Mingi Kim
Chung-Ang University
7 shared publications- Ji Woo Hong
Korea Advanced Institute of Science and Technology
6 shared publications - Jongwon Ryu
AI Graduate School, Chung-Ang University, Seoul, Repulic of Korea
6 shared publications - Minuk Ma
University of British Columbia
6 shared publications
Latest Publications
- 6 August 2026
- Published by Springer Nature
- p. 76-90
The Pragmatic Persona: Discovering LLM Persona Through Bridging Inference
- 6 August 2026
- Published by Springer Nature
- p. 76-90
The Pragmatic Persona: Discovering LLM Persona Through Bridging Inference
- 3 August 2026
- Published by Springer Nature
- p. 669-684
Blocking Visual Leakage: Visually-Agnostic Text Decomposition for Composed Video Retrieval
- 3 August 2026
- Published by Springer Nature
- p. 669-684
Blocking Visual Leakage: Visually-Agnostic Text Decomposition for Composed Video Retrieval
- 3 August 2026
- Published by Springer Nature
- p. 523-537
Video DETOX: Purifying Noisy Relevance Signals for Diverse and Long-Form Video Understanding
- 3 August 2026
- Published by Springer Nature
- p. 523-537
Video DETOX: Purifying Noisy Relevance Signals for Diverse and Long-Form Video Understanding
- 3 August 2026
- Published by Springer Nature
- p. 383-396
Beyond Co-existence: Measuring Attribute Binding Hallucinations in Audio-Language Models
- 3 August 2026
- Published by Springer Nature
- p. 383-396
Beyond Co-existence: Measuring Attribute Binding Hallucinations in Audio-Language Models
- Open Access
- 16 June 2026
- Published by Springer Nature in BMC Emergency Medicine
- Vol. 26 (1) , 173
Deep learning-based natural language processing for critical care identification in pediatric emergency department
The initial assessment of pediatric emergency patients is challenging due to diverse clinical presentations and the limited applicability of existing morta
- 16 June 2026
- Published by Springer Nature in BMC Emergency Medicine
- Vol. 26 (1) , 173
Open AccessDeep learning-based natural language processing for critical care identification in pediatric emergency department
The initial assessment of pediatric emergency patients is challenging due to diverse clinical presentations and the limited applicability of existing morta
- Open Access
- 2 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 51576-51588
SLICE: An Efficient and Tuning-Free Keyframe Sampling Framework for Long-Form Video Understanding
Efficiently distilling long videos into semantically significant keyframes is essential for effective Video Question Answering (VideoQA) using Large Multimodal Models (LMMs). However, existing sampling strategies suffer from hyperparameter brittleness, relying heavily on manual hyperparameter tuning or dataset-specific calibration. To address these limitations, we introduce SLICE (Semantic Length-Independent Content Extractor), a parameter-free framework. The core mechanism of SLICE introduces logarithmic scale-adaptive smoothing (σ = ln(N)) based on the video length N to suppress visual noise, and logically allocates a limited token budget through Information Density Partitioning. By partitioning videos based on actual Semantic Energy Mass rather than uniform time intervals, SLICE concentrates frames within complex event segments and aggressively omits static backgrounds, thereby securing both temporal diversity and semantic relevance. Experimental results demonstrate that SLICE significantly outperforms hyperparameter-sensitive state-of-the-art (SOTA) methods. Notably, on sparse event localization tasks, SLICE achieves an outstanding visual F1 score of up to 80.2% on HAYSTACK-LVBENCH and dramatically improves the temporal F1 score on HAYSTACK-EGO4D by nearly 3 times compared to previous baselines, all while reducing computational overhead. These findings prove that SLICE is a highly scalable, out-of-the-box solution for real-world LMM deployments.
- 2 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 51576-51588
Open AccessSLICE: An Efficient and Tuning-Free Keyframe Sampling Framework for Long-Form Video Understanding
Efficiently distilling long videos into semantically significant keyframes is essential for effective Video Question Answering (VideoQA) using Large Multimodal Models (LMMs). However, existing sampling strategies suffer from hyperparameter brittleness, relying heavily on manual hyperparameter tuning or dataset-specific calibration. To address these limitations, we introduce SLICE (Semantic Length-Independent Content Extractor), a parameter-free framework. The core mechanism of SLICE introduces logarithmic scale-adaptive smoothing (σ = ln(N)) based on the video length N to suppress visual noise, and logically allocates a limited token budget through Information Density Partitioning. By partitioning videos based on actual Semantic Energy Mass rather than uniform time intervals, SLICE concentrates frames within complex event segments and aggressively omits static backgrounds, thereby securing both temporal diversity and semantic relevance. Experimental results demonstrate that SLICE significantly outperforms hyperparameter-sensitive state-of-the-art (SOTA) methods. Notably, on sparse event localization tasks, SLICE achieves an outstanding visual F1 score of up to 80.2% on HAYSTACK-LVBENCH and dramatically improves the temporal F1 score on HAYSTACK-EGO4D by nearly 3 times compared to previous baselines, all while reducing computational overhead. These findings prove that SLICE is a highly scalable, out-of-the-box solution for real-world LMM deployments.
- Open Access
- 23 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 29984-29994
Lexical Distractor Mining Network for Causal Video Question Answering
Causal Video Question Answering (Causal VideoQA) aims to answer questions about a video by reasoning over causal relationships among multiple events within the video. Unlike conventional VideoQA, which involves selecting a single correct answer from multiple choices, Causal VideoQA requires selecting both the answer and the causal reasoning that supports it. Despite significant progress, recent Causal VideoQA systems often struggle to identify accurate causal reasoning. This is largely due to the presence of lexical distractors in the reasoning candidates, which confuse models and lead them to incorrect inferences. A lexical distractor refers to an incorrect reasoning choice that mimics the correct reasoning sentence by exploiting the lexical and structural similarity of words, without reflecting the actual intended meaning. To this end, we propose the Lexical Distractor Mining Network (LDMNet), which performs sensible causal reasoning by progressively mining these distractors using our defined lexical similarity score and utilizing them as hard negatives for contrastive learning. Extensive experiments on two recent Causal VideoQA benchmarks (i.e., Causal-VidQA, NExT-QA) demonstrate that LDMNet achieves state-of-the-art performance, surpassing existing methods across multiple question types by reducing the model’s unnecessary reliance on lexical cues. The code will be made publicly available.
- 23 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 29984-29994
Open AccessLexical Distractor Mining Network for Causal Video Question Answering
Causal Video Question Answering (Causal VideoQA) aims to answer questions about a video by reasoning over causal relationships among multiple events within the video. Unlike conventional VideoQA, which involves selecting a single correct answer from multiple choices, Causal VideoQA requires selecting both the answer and the causal reasoning that supports it. Despite significant progress, recent Causal VideoQA systems often struggle to identify accurate causal reasoning. This is largely due to the presence of lexical distractors in the reasoning candidates, which confuse models and lead them to incorrect inferences. A lexical distractor refers to an incorrect reasoning choice that mimics the correct reasoning sentence by exploiting the lexical and structural similarity of words, without reflecting the actual intended meaning. To this end, we propose the Lexical Distractor Mining Network (LDMNet), which performs sensible causal reasoning by progressively mining these distractors using our defined lexical similarity score and utilizing them as hard negatives for contrastive learning. Extensive experiments on two recent Causal VideoQA benchmarks (i.e., Causal-VidQA, NExT-QA) demonstrate that LDMNet achieves state-of-the-art performance, surpassing existing methods across multiple question types by reducing the model’s unnecessary reliance on lexical cues. The code will be made publicly available.
- Open Access
See More, Store Less: Memory-Efficient Resolution for Video Moment Retrieval
Recent advances in Multimodal Large Language Models (MLLMs) have improved image recognition and reasoning, but video-related tasks remain challenging due to memory constraints from dense frame processing. Existing Video Moment Retrieval (VMR) methodologies rely on sparse frame sampling, risking potential information loss, especially in lengthy videos. We propose SMORE (See MORE, store less), a framework that enhances memory efficiency while maintaining high information resolution. SMORE (1) uses query-guided captions to encode semantics aligned with user intent, (2) applies query-aware importance modulation to highlight relevant segments, and (3) adaptively compresses frames to preserve key content while reducing redundancy. This enables efficient video understanding without exceeding memory budgets. Experimental validation reveals that SMORE achieves state-of-the-art performance on QVHighlights, Charades-STA, and ActivityNet-Captions benchmarks.
Open AccessSee More, Store Less: Memory-Efficient Resolution for Video Moment Retrieval
Recent advances in Multimodal Large Language Models (MLLMs) have improved image recognition and reasoning, but video-related tasks remain challenging due to memory constraints from dense frame processing. Existing Video Moment Retrieval (VMR) methodologies rely on sparse frame sampling, risking potential information loss, especially in lengthy videos. We propose SMORE (See MORE, store less), a framework that enhances memory efficiency while maintaining high information resolution. SMORE (1) uses query-guided captions to encode semantics aligned with user intent, (2) applies query-aware importance modulation to highlight relevant segments, and (3) adaptively compresses frames to preserve key content while reducing redundancy. This enables efficient video understanding without exceeding memory budgets. Experimental validation reveals that SMORE achieves state-of-the-art performance on QVHighlights, Charades-STA, and ActivityNet-Captions benchmarks.
- Open Access
Language-Grounded Multi-Domain Image Translation via Semantic Difference Guidance
Multi-domain image-to-image translation re quires grounding semantic differences ex pressed in natural language prompts into corresponding visual transformations, while preserving unrelated structural and seman tic content. Existing methods struggle to maintain structural integrity and provide fine grained, attribute-specific control, especially when multiple domains are involved. We propose LACE (Language-grounded Attribute Controllable Translation), built on two compo nents: (1) a GLIP-Adapter that fuses global semantics with local structural features to pre serve consistency, and (2) a Multi-Domain Control Guidance mechanism that explicitly grounds the semantic delta between source and target prompts into per-attribute translation vec tors, aligning linguistic semantics with domain level visual changes. Together, these modules enable compositional multi-domain control with independent strength modulation for each attribute. Experiments on CelebA(Dialog) and BDD100K demonstrate that LACE achieves high visual fidelity, structural preservation, and interpretable domain-specific control, surpass ing prior baselines. This positions LACE as a cross-modal content generation framework bridging language semantics and controllable visual translation.
Open AccessLanguage-Grounded Multi-Domain Image Translation via Semantic Difference Guidance
Multi-domain image-to-image translation re quires grounding semantic differences ex pressed in natural language prompts into corresponding visual transformations, while preserving unrelated structural and seman tic content. Existing methods struggle to maintain structural integrity and provide fine grained, attribute-specific control, especially when multiple domains are involved. We propose LACE (Language-grounded Attribute Controllable Translation), built on two compo nents: (1) a GLIP-Adapter that fuses global semantics with local structural features to pre serve consistency, and (2) a Multi-Domain Control Guidance mechanism that explicitly grounds the semantic delta between source and target prompts into per-attribute translation vec tors, aligning linguistic semantics with domain level visual changes. Together, these modules enable compositional multi-domain control with independent strength modulation for each attribute. Experiments on CelebA(Dialog) and BDD100K demonstrate that LACE achieves high visual fidelity, structural preservation, and interpretable domain-specific control, surpass ing prior baselines. This positions LACE as a cross-modal content generation framework bridging language semantics and controllable visual translation.
- Open Access
GranAlign: Granularity-Aware Alignment Framework for Zero-Shot Video Moment Retrieval
Zero-shot video moment retrieval (ZVMR) is the task of localizing a temporal moment within an untrimmed video using a natural language query without relying on task-specific training data. The primary challenge in this setting lies in the mismatch in semantic granularity between textual queries and visual content. Previous studies in ZVMR have attempted to achieve alignment by leveraging high-quality pre-trained knowledge that represents video and language in a joint space. However, these approaches failed to balance the semantic granularity between the pre-trained knowledge provided by each modality for a given scene. As a result, despite the high quality of each modality's representations, the mismatch in granularity led to inaccurate retrieval. In this paper, we propose a training-free framework, called Granularity-Aware Alignment (GranAlign), that bridges this gap between coarse and fine semantic representations. Our approach introduces two complementary techniques: granularity-based query rewriting to generate varied semantic granularities, and query-aware caption generation to embed query intent into video content. By pairing multi-level queries with both query-agnostic and query-aware captions, we effectively resolve semantic mismatches. As a result, our method sets a new state-of-the-art across all three major benchmarks (QVHighlights, Charades-STA, ActivityNet-Captions), with a notable 3.23% mAP@avg improvement on the challenging QVHighlights dataset.
Open AccessGranAlign: Granularity-Aware Alignment Framework for Zero-Shot Video Moment Retrieval
Zero-shot video moment retrieval (ZVMR) is the task of localizing a temporal moment within an untrimmed video using a natural language query without relying on task-specific training data. The primary challenge in this setting lies in the mismatch in semantic granularity between textual queries and visual content. Previous studies in ZVMR have attempted to achieve alignment by leveraging high-quality pre-trained knowledge that represents video and language in a joint space. However, these approaches failed to balance the semantic granularity between the pre-trained knowledge provided by each modality for a given scene. As a result, despite the high quality of each modality's representations, the mismatch in granularity led to inaccurate retrieval. In this paper, we propose a training-free framework, called Granularity-Aware Alignment (GranAlign), that bridges this gap between coarse and fine semantic representations. Our approach introduces two complementary techniques: granularity-based query rewriting to generate varied semantic granularities, and query-aware caption generation to embed query intent into video content. By pairing multi-level queries with both query-agnostic and query-aware captions, we effectively resolve semantic mismatches. As a result, our method sets a new state-of-the-art across all three major benchmarks (QVHighlights, Charades-STA, ActivityNet-Captions), with a notable 3.23% mAP@avg improvement on the challenging QVHighlights dataset.