Research map / Video / Video-language understanding
Long Story Short: Story-level Video Understanding from 20K Short Films
Ridouane Ghermi, Xi Wang, Vicky S. Kalogeiton, Ivan Laptev
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 |
|---|---|---|---|
| give the movie title to the LLM (ablation) | GPT-4 title-only · MovieQA | accuracy: 53.9 -> 71.3 | Improved |
| add character names (no anonymization) on top of title (ablation) | GPT-4 title-only · MovieQA | accuracy: 31.4 -> 53.9 | Improved |
| give the movie title to the LLM (ablation) | GPT-4 title-only · SF20K-Test | accuracy: 17.1 -> 18.3 | No clear effect |
| add character names to the question (ablation) | GPT-4 title-only · SF20K-Test | accuracy: 15.9 -> 17.1 | No clear effect |
| replace the title with a random commercial title | LLM title-only · MovieQA | accuracy: 36.3 -> 20.3 | Hurt |
| replace the title with a random commercial title | LLM title-only · SF20K-Test | accuracy: 17.1 -> 8.5 | Hurt |
| replace the title with a random amateur title | LLM title-only · MovieQA | accuracy: 36.3 -> 32.0 | Hurt |
| replace the title with a random amateur title | LLM title-only · SF20K-Test | accuracy: 17.1 -> 16.7 | No clear effect |
| add subtitles and video frames to a single modality (ablation) | GPT-5 · SF20K-Test | accuracy: 28.6 -> 56.2 | Improved |
| add video frames to language-only input (ablation) | GPT-5 · SF20K-Test | accuracy: 37.7 -> 56.2 | Improved |
| increase frames from one to 128 (ablation) | GPT-5 · SF20K-Test | accuracy: 39.6 -> 58.6 | Improved |
| enlarge temporal window from shot to scene (ablation) | Qwen3-VL-8B · SF20K-Test | accuracy: 22.3 -> 25.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 2 on most shared results.
- LongVU-7B — LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding (ICML 2025) · better on 0 of 1 shared result
- LVAgent — LVAgent: Long Video Understanding by Multi-Round Dynamical Collaboration of MLLM Agents (ICCV 2025) · better on 0 of 1 shared result
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
- Unleashing Hour-Scale Video Training for Long Video-Language Understanding (NeurIPS 2025 · similarity 0.90)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.90)
- HourVideo: 1-Hour Video-Language Understanding (NeurIPS 2024 · similarity 0.89)
- Leveraging Textual Memory and Key Frame Reasoning for Full Video Understanding Using Off-the-Shelf LLMs and VLMs (Student Abstract) (AAAI 2025 · similarity 0.89)
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.89)
- SF2T: Self-supervised Fragment Finetuning of Video-LLMs for Fine-Grained Understanding (CVPR 2025 · similarity 0.88)
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.
