Long video understanding requires models to identify and reason over sparse, temporally distributed evidence within limited frame budgets. Existing keyframe selection methods typically retrieve query-relevant frames from a densely sampled candidate set before inference. However, global image-text matching incurs unnecessary computation when a small set of uniformly sampled frames already provides sufficient evidence. Query relevance alone may also overlook evidence that distinguishes competing answers, while one-shot selection offers no opportunity to correct earlier selection errors. We propose HypoKFS, a training-free, plug-and-play framework that couples keyframe selection with iterative inference by multimodal large language models (MLLMs). HypoKFS first reasons over sparsely sampled frames to construct an initial hypothesis space. The resulting answer distribution determines whether additional frame selection is needed. When needed, the retained hypotheses guide the selection of discriminative frames, followed by renewed inference and hypothesis refinement. This iterative process directs additional visual computation toward unresolved hypotheses while avoiding unnecessary global frame matching. Experiments on Video-MME, LongVideoBench, and MLVU demonstrate consistent improvements across multiple MLLM backbones. HypoKFS improves accuracy by up to 6.7 percentage points and achieves 72.9% accuracy on Video-MME at only 1.3% of the reference frame-hypothesis matching cost.
Figures & tables
Figure 1: Evolution of visual evidence selection for VideoQA. (a) Uniform sampling uses fixed, query-agnostic frames. (b) Conventional KFS performs global image–text matching and frame selection before MLLM inference. (c) HypoKFS first infers from sparsely sampled frames and uses the resulting hypothesis space to guide subsequent frame selection.
Figure 2: Overview of HypoKFS. Initial inference from sparse frames constructs a compact hypothesis space and determines whether additional computation is required. HypoKFS then iteratively selects discriminative frames and updates the hypothesis space through renewed inference for at most M rounds.
Model
Frames
Params
LVB
MLVU
V-MME
Video MLLMs
GPT-5 OpenAI (2025)
–
–
72.6
77.3
81.8
Gemini-2.5-Pro Gemini Team, Google (2025)
–
–
77.6
–
84.8
Qwen2.5-VL-72B Bai et al. (2025)
256
72B
60.7
74.6
73.3
InternVL3-8B Zhu et al. (2025)
32
8B
62.3
71.4
66.5
ViLAMP Cheng et al. (2025)
1fps
7B
67.5
–
61.2
Table 1: Performance comparison on LongVideoBench (LVB), MLVU, and Video-MME (V-MME). For HypoKFS, Frames / Budget reports the average total number of processed frames and the per-inference frame budget, respectively.
Figure 3: Analysis of hypothesis-guided selective retrieval on Video-MME. (a) Accuracy–retrieval efficiency. (b) Cumulative top- K hypothesis coverage. (c) Accuracy–computation trade-off under different thresholds τ .
Table 2: Ablation studies on Video-MME. (a) Component ablation across video durations. (b) Hyperparameter sensitivity to α and λ under keyframe budgets K=64/128/256 , where each cell reports the corresponding accuracies in this order.
Figure 4: Qualitative comparison on Video-MME. Uniform sampling may miss sparse evidence, while relevance-based retrieval can favor query-related yet non-discriminative frames. HypoKFS instead retrieves evidence that resolves competing answer hypotheses.
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Model
Frames
Params
Video-MME
Qwen3-VL
128
8B
69.1
+ HypoKFS
128
8B
72.9
Qwen3-VL
256
8B
70.2
+ HypoKFS
256
8B
73.8
Qwen3-VL
1 fps
8B
70.8
+ HypoKFS
1 fps
8B
74.3
Appendix
Table 3: Scaling performance under different frame settings on Video-MME using Qwen3-VL.
Long-video understanding remains challenging for multimodal large language models (MLLMs) because densely encoding long frame sequences is computationally expensive, while uniform sampling under a limited visual budget can miss sparse yet decisive evidence. Recent training-free keyframe selection methods have enabled more efficient inference and yielded promising performance gains. However, many existing methods score frames largely in isolation without explicitly considering how each candidate complements the currently selected subset, potentially resulting in redundant selections and incomplete evidence coverage. To address this limitation, we propose MarKey, a training-free framework that formulates keyframe selection as subset-aware greedy optimization. At each iteration, MarKey scores each candidate using a tractable surrogate that jointly accounts for query relevance, marginal coverage gain, and context-dependent redundancy, and selects the frame with the highest utility. To make this iterative subset-aware evaluation efficient, MarKey uses a compact set of representative anchors to approximate full-video coverage and a bounded window of previously selected frames to limit context-dependent comparisons. Experiments on six benchmarks spanning holistic video understanding, human-centric video understanding, and open-ended video understanding demonstrate that MarKey consistently outperforms existing methods. Further analyses show robust gains across different MLLM backbones, model scales, and frame budgets.
Video understanding is often plagued by severe temporal redundancy, where processing dense frame sequences is both semantically inefficient and computationally expensive. This challenge is further amplified when only a small subset of frames is truly relevant to the given query. In this paper, we propose a Query- and Content-Aware (QCA) keyframe selection framework that can select a compact yet information-rich set of frames from long videos. QCA first partitions the video into temporal segments and estimates the information contribution of each segment by jointly modeling query relevance and content deviation, and dynamically allocates keyframe budget to each segment. Within each segment, QCA anchors on the most query-relevant frame and iteratively incorporates additional frames to maximize diversity while maintaining high semantic relevance to the query. Crucially, our method requires no additional training and can be seamlessly integrated into existing Video-LLMs. Extensive experiments across multiple long video understanding benchmarks demonstrate that our proposed approach achieves state-of-the-art performance and has strong generalization ability. For instance, QCA achieves 67.8% on LongVideoBench using 128 frames, while GPT-4o achieves 66.7% using 256 frames. Our codes are available in GitHub.
Jun Peng, Baiyang Song, Jie Li +4
Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, P.R. China · Peng Cheng Laboratory, Shenzhen, P.R. China · School of Electronic and Computer Engineering, Peking University, P.R. China
Multimodal large language models (MLLMs) process multimodal inputs by converting text, images, and videos into token sequences that are subsequently processed by a backbone language model. While MLLMs have achieved excellent performance in understanding the content of individual images, video understanding remains significantly more difficult because videos contain large number of video frames. MLLMs typically process only a subset of these frames, usually ranging from 8 to 64. MLLMs usually sample frames uniformly, regardless of their relevance to the question being answered. To address this limitation, several training-free, model-agnostic methods for selecting question-relevant frames have recently been proposed. In this work, we introduce a frame-selection method designed specifically for multiple-choice questions. We extend the query text by appending semantic cues derived from the answer choices and employ a direct query-frame alignment scoring mechanism. To the best of our knowledge, our method is the first to directly utilize answer choices as inference-time cues for selecting frames relevant to answering a question. The method first constructs a compact candidate pool by subsampling video frames at a fixed rate. The frames are then scored according to their maximum cosine similarity across all question-answer pairs to identify the most relevant frames for a given query. This approach preserves a fixed token budget while improving the relevance of the visual evidence provided to the downstream MLLM. We evaluate the effectiveness of our frame-selection method on the MLVU, Video-MME, and LongVideoBench benchmarks using three MLLMs: LLaVA-Mini, Qwen2-VL, and LLaVA-Video. Experimental results demonstrate that answer-aware frame selection generally outperforms uniform sampling and existing training-free frame-selection methods under the same frame budget.
Md. Safayet Islam, Dilip Sarkar, Liang Liang
Department of Computer Science University of Miami Coral Gables, FL 33146