47
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
0.0606
0.2001
0.5949
0.1913
0.0146
0.0344
Qwen3-VL-30B
0.3230
0.0715
0.2154
0.6053
0.1958
0.0224
0.0544
RVLSM (ours)
Qwen3-Omni-30B
w/ STD
0.3035
0.0584
0.2011
0.5911
0.1806
0.0143
0.0681
+MTC
0.2929
0.0497
0.1962
0.5840
0.1552
0.0108
0.0506
+AMTC
0.2815
0.0449
0.1929
0.6010
0.1912
0.0121
0.0356
Qwen3-VL-30B
w/ STD
0.3326
0.0745
0.2257
0.6380
0.2023
0.0176
0.0739
+MTC
0.3024
0.0571
0.2077
0.6094
0.1589
0.0124
0.0714
+AMTC
0.3100
0.0595
0.2031
0.6134
0.1873
0.0132
0.0532
Table 1: Performance comparison between RVLSM (ours), the traditional (non-recursive) VideoLLM baselines, and the recently proposed V2Xum model (Hua et al. 2025) fine-tuned on VideoXum. The RVLSM block reports the recursive pipeline with scene-aware STD, and its combination with multimodal tool calling (STD+MTC) and adaptive multimodal tool calling (STD+AMTC), across two representative Sub-VLLMs.
Figure 2: Proprietary and open-source Root-LLM: performance gap between GPT-5.4 and Qwen-480B across datasets and settings (w/ and w/o STD).
(a)
(b)
Figure 3: SBERT comparison on MixTyVSum and VideoXum (a) and across the MixTyVSum subsets (b).
Ablation study (RQ2): what are
the effects of the main RVLSMs’s components?
We examine the role of Adaptive Multimodal Tool Calling (AMTC) by comparing three scenarios: RVLSM+STD, which employs scene-aware decomposition without multimodal tool integration; RVLSM+MTC, which uses multimodal tools but not adaptively; and RVLSM+AMTC, which employs multimodal tools adaptively. Table 1 reports the performance of the two sub-VLLMs under these strategies, and Figure 3(a) summarizes the corresponding SBERT behavior on the two benchmarks.
A clear divergence arises between the benchmarks due to differing annotation styles. On MixTyVSum, which features action-oriented and semantically dense summaries, adaptive injection enhances semantic quality (SBERT and METEOR).
However, this can lead to a decline in ROUGE scores, particularly for Qwen3-Omni-30B, which performs best with STD alone, indicating a trade-off between semantic detail and lexical overlap. Conversely, on VideoXum, where summaries are more generic, injection is largely ineffective, with RVLSM+STD achieving the best scores across most metrics for both sub-VLLMs.
Additionally, Figure 3(b) decomposes SBERT by source dataset and shows that the benefit of injection depends on how well STD alone already captures the video content. On structured, procedural subsets such as EpicKitchens and SumMe, STD is already strong and injection yields no gain or a marginal drop; on more diverse subsets such as TVSum and ActivityNet, adaptive injection provides substantial improvements (TVSum: STD=0.551 to AMTC=0.701; ActivityNet: STD=0.575 to AMTC=0.655). Across all four subsets, RVLSM+AMTC consistently outperforms RVLSM+MTC, and its advantage is largest precisely in the structured domains where indiscriminate injection is most harmful (EpicKitchens: MTC=0.577 to AMTC=0.600; SumMe: MTC=0.577 to AMTC=0.611). This confirms the benefit of adaptive tool selection over unconditional invocation: applying every multimodal analysis regardless of context introduces noise that offsets, and in structured domains, outweighs the benefit of the injected signal.
Semantic alignment (RQ3): are RVLSMs’s summaries closely aligned with human-generated ones?
We empirically verify whether the RVLSM’s summaries are more closely aligned with human expectations
than baseline methods in terms of NLP entities retrieved in the ground-truth summary.
The core idea is to verify the semantic alignment in terms of matching with semantically ric