Research map / Video / Video-language understanding
MERLOT: Multimodal Neural Script Knowledge Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Pretrain on videos rather than static image-caption pairs (ablation) | MERLOT · VCR validation Q→A | accuracy: 58.9 -> 75.2 | Helped |
| Use diverse YT-Temporal-180M rather than instructional-only HowTo100M data (ablation) | MERLOT · VCR validation Q→A | accuracy: 66.3 -> 72.8 | Helped |
| Use the full YT-Temporal-180M corpus rather than its HowTo100M-sized subset (ablation) | MERLOT · VCR validation Q→A | accuracy: 72.8 -> 75.2 | Helped |
| Use denoised ASR rather than raw ASR (ablation) | MERLOT · VCR validation Q→A | accuracy: 72.8 -> 75.2 | Helped |
| Train longer (ablation) | MERLOT · VCR validation Q→A | accuracy: 75.2 -> 75.9 -> 77.0 -> 78.5 -> 79.4 | Helped |
| Add attention masking to four-segment context (ablation) | MERLOT · VCR validation Q→A | accuracy: 74.1 -> 75.2 | Helped |
| Add attention masking to four-segment context (ablation) | MERLOT · TVQA+ | accuracy: 73.3 -> 75.8 | Helped |
| Add attention masking with one-segment context (ablation) | MERLOT · VCR validation Q→A | accuracy: 73.8 -> 73.5 | No clear effect |
| Add contrastive frame-transcript matching loss (ablation) | MERLOT · VCR validation Q→A | accuracy: 57.5 -> 75.2 | Helped |
| Add contrastive frame-transcript matching loss (ablation) | MERLOT · TVQA+ | accuracy: 67.6 -> 75.8 | Helped |
| Add temporal ordering loss (ablation) | MERLOT · TVQA+ | accuracy: 75.6 -> 75.8 | Helped |
| Add temporal ordering loss (ablation) | MERLOT · VCR validation Q→A | accuracy: 75.5 -> 75.2 | Hurt |
Closest papers by meaning
- MULTISCRIPT: Multimodal Script Learning for Supporting Open Domain Everyday Tasks (AAAI 2024 · similarity 0.88)
- MMOU: A Massive Multi-Task Omni Understanding and Reasoning Benchmark for Long and Complex Real-World Videos (NeurIPS 2026 · similarity 0.88)
- SciVideoBench: Benchmarking Scientific Video Reasoning in Large Multimodal Models (ICML 2026 · similarity 0.88)
- MVBench: A Comprehensive Multi-modal Video Understanding Benchmark (CVPR 2024 · similarity 0.88)
- TOMATO: Assessing Visual Temporal Reasoning Capabilities in Multimodal Foundation Models (ICLR 2025 · similarity 0.88)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.87)
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.