Research map / Video / Video-language understanding
Agentic Video Summarization via Self-Reflecting Multimodal Understanding
Miaotian Guo, Shuguang Dou, Yin Li, Aidong Men, Dongsheng Jiang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add V2I alignment of video and image embeddings in the Summarizer (ablation) | CSTA baseline · SumMe | Kendall tau: 0.230 -> 0.265 | Improved |
| add V2I alignment (ablation) | CSTA baseline · TVSum | Spearman rho: 0.232 -> 0.278 | Improved |
| add Verifier and Reflector on top of V2I Summarizer (ablation) | AgenticVS · SumMe | Kendall tau: 0.265 -> 0.274 | Improved |
| add Verifier and Reflector on top of V2I Summarizer (ablation) | AgenticVS · TVSum | Kendall tau: 0.215 -> 0.220 | Improved |
| align video to image space (V2I) instead of image to video (I2V) (ablation) | V2I Alignment · SumMe | Kendall tau: 0.238 -> 0.265 | Improved |
| use temporal attention pooling instead of mean pooling (ablation) | V2I Alignment · SumMe | Kendall tau: 0.243 -> 0.265 | Improved |
| fuse image and video encoders with alignment instead of using either alone (ablation) | Summarizer · SumMe | Kendall tau: 0.228 (image only), 0.220 (video only) -> 0.265 | Improved |
| directly concatenate image and video embeddings instead of aligning (ablation) | Summarizer · SumMe | Kendall tau: 0.228 -> 0.230 | No clear effect |
| use CLIP similarity to MLLM-generated summary caption instead of MLLM direct scoring (ablation) | Reflector · SumMe (train-free) | Kendall tau: 0.073 -> 0.116 | Improved |
| use AgenticVS instead of LLMVS | LLMVS · SumMe | Kendall tau: 0.253 -> 0.274 | Improved |
| use Verifier without Reflector (ablation) | AgenticVS · SumMe and TVSum | Kendall tau | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CSTA — CSTA: CNN-based Spatiotemporal Attention for Video Summarization (CVPR 2024) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- HAS: Highlight-guided Attention Steering for Multimodal LLM Video Summarization (ECCV 2026 · similarity 0.87)
- AgenticVBench: Can AI Agents Complete Real-World Video Production Tasks? (NeurIPS 2026 · similarity 0.87)
- Video Summarization with Large Language Models (CVPR 2025 · similarity 0.86)
- VELOCITI: Benchmarking Video-Language Compositional Reasoning with Strict Entailment (CVPR 2025 · similarity 0.85)
- Deep Video Discovery: Agentic Search with Tool Use for Long-form Video Understanding (NeurIPS 2025 · similarity 0.85)
- V2Xum-LLM: Cross-Modal Video Summarization with Temporal Prompt Instruction Tuning (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
