n in the Supplementary Document.
Auxiliary Modules
For Named Entity Recognition we use SpaCy (Honnibal et al. 2020) (en_core_web_sm model).
For scene-aware STD we use PySceneDetect (Castellano 2014).
Object-level semantic signals are extracted using YOLOv8 (Jocher and Qiu 2023) pretrained on the COCO dataset, applied every 10 frames.
Action-level semantic evidence is obtained through MMAction2 (MMAction2 Contributors 2020) using a TSN model (ResNet-50, Kinetics-400), retaining only predictions above a confidence threshold of 0.800.80.
Automatic speech recognition is extracted using Whisper (Radford et al. 2022) (medium variant), and optical character recognition is detected using EasyOCR (AI 2020).
Main Results and Discussion
Peformance comparison (RQ1): Is RLM beneficial for LLM-based video-to-text summarization of long-form videos compared to non-recursive zero-shot baselines?
Table 1 reports the main summarization results on MixTyVSum and VideoXum across the different sub-VLLM settings.
Recursive video-to-text summarization outperforms the monolithic zero-shot baselines, demonstrating the benefit of STD for long videos. In fact, on MixTyVSum, the strongest performance is obtained by RVLSM with STD paired with Qwen3-Omni-30B, which achieves the highest ROUGE-1, ROUGE-2, and ROUGE-L scores; on VideoXum, RVLSM with Qwen3-VL-30B yields the best results overall, attaining both the highest ROUGE scores and the highest SBERT similarity. METEOR follows the same ordering as ROUGE across both benchmarks, corroborating this ranking.
The comparison with the fine-tuned V2Xum-LLaMA-7B (Hua et al. 2025) is informative for assessing generalization. On VideoXum, on which it was optimized, the fine-tuned model retains an advantage on surface-overlap metrics, yet the best zero-shot RVLSM configuration still surpasses it on ROUGE-1, METEOR, and, most notably, on SBERT similarity (0.6380 vs. 0.5573), indicating that recursive summarization captures the semantics of the reference summaries even without task-specific training. On MixTyVSum the picture reverses: the fine-tuned model falls below the zero-shot sub-VLLMs on nearly all metrics, whereas RVLSM improves over both baselines by a wide margin. This suggests that a model tuned to a single summarization distribution transfers poorly across benchmarks, while the recursive approach generalizes robustly.
Choice of the Root-LLM: Proprietary vs. Opensource
The choice of Root-LLM markedly affects recursive summarization quality. Figure 2 shows, for each combination of sub-VLLM and metric, the differences in performance scores between using GPT-5.4 and Qwen-480B as Root-LLM, under both fixed-duration (w/o STD) and scene-aware (w/ STD) decomposition. Results show that the peformance is roughly comparable, underscoring the practical advantages of using opensource tools.
Sub-VLLM
Method
R1F1 ↑\uparrow
R2F1 ↑\uparrow
RLF1 ↑\uparrow
SBERT ↑\uparrow
METEOR ↑\uparrow
BLEU-4 ↑\uparrow
CIDEr ↑\uparrow
MixTyVSum dataset
Fine-Tuned Baseline
V2Xum-LLaMA-7B (Fine-Tuned)
0.2115
0.0392
0.1603
0.4463
0.1092
0.0043
0.0152
Zero-Shot Baselines
Molmo2-8B
0.3212
0.0694
0.2018
0.5445
0.1416
0.0124
0.0260
Qwen3-VL-8B
0.2669
0.0504
0.1797
0.5664
0.1245
0.0169
0.0474
Qwen3-Omni-30B
0.2946
0.0682
0.1947
0.5788
0.1624
0.0137
0.0435
Qwen3-VL-30B
0.3132
0.0736
0.2057
0.5794
0.1538
0.0146
0.0459
RVLSM (ours)
Qwen3-Omni-30B
w/ STD
0.3606
0.0920
0.2298
0.5859
0.1527
0.0149
0.0542
+MTC
0.3233
0.0719
0.2083
0.5778
0.1737
0.0130
0.0536
+AMTC
0.3068
0.0643
0.1924
0.6063
0.1652
0.0120
0.0570
Qwen3-VL-30B
w/ STD
0.3234
0.0753
0.2113
0.5958
0.1547
0.0121
0.0513
+MTC
0.3303
0.0803
0.2142
0.6149
0.1724
0.0140
0.0568
+AMTC
0.3192
0.0683
0.2051
0.6401
0.1801
0.0146
0.0531
VideoXum dataset
Fine-Tuned Baseline
V2Xum-LLaMA-7B (F.T.)
0.3228
0.1016
0.2475
0.5573
0.1826
0.0292
0.1386
Zero-Shot Baselines
Molmo2-8B
0.2860
0.0651
0.1843
0.5975
0.1920
0.0142
0.0208
Qwen3-VL-8B
0.2761
0.0428
0.1840
0.5799
0.1558
0.0122
0.0478
Qwen3-Omni-30B
0.3041
0.06