Research map / Video / Video-language understanding
LGDN: Language-Guided Denoising Network for Video-Language Modeling
Haoyu Lu, Mingyu Ding, Nanyi Fei, Yuqi Huo, Zhiwu Lu
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 |
|---|---|---|---|
| Use language-guided SFP instead of selecting frames without SFP (ablation) | LGDN · MSR-VTT 1k-A test | R@SUM: 336.1 -> 352.9 | Improved |
| Add MFCL with MSL instead of standard frame-level contrastive learning (ablation) | LGDN · MSR-VTT 1k-A test | R@SUM: 330.5 -> 335.2 | Improved |
| Add MFCL without MSL (ablation) | LGDN without SFP · MSR-VTT 1k-A test | R@SUM: 331.2 -> 330.5 | Hurt |
| Add MVCL video-level contrastive learning (ablation) | LGDN with LSFM and MFCL, without SFP · MSR-VTT 1k-A test | R@SUM: 335.2 -> 336.1 | Improved |
| Use ensemble global and local alignment at inference instead of local alignment alone (ablation) | LGDN · MSR-VTT 1k-A test | R@SUM: 352.9 -> 360.4 | Improved |
| Use global alignment only at inference instead of local alignment (ablation) | LGDN · MSR-VTT 1k-A test | R@SUM: 352.9 -> 330.7 | Hurt |
| Combine Momentum and CrossMom relevance scores (ablation) | LGDN SFP relevance estimator · MSR-VTT 1k-A test | R@SUM: 359.7 -> 360.4 | Improved |
| Use momentum encoder supervision for relevance scoring instead of SimDot (ablation) | LGDN SFP relevance estimator · MSR-VTT 1k-A test | R@SUM: 356.5 -> 359.5 | Improved |
| Increase memory bank size from 1,200 to 9,600 (ablation) | LGDN · MSR-VTT 1k-A test | Text-to-video R@1: 36.0 -> 38.9 | Improved |
| Increase memory bank size from 9,600 to 19,200 (ablation) | LGDN · MSR-VTT 1k-A test | Text-to-video R@1: 38.9 -> 38.3 | Hurt |
| Add SFP frame selection | Clip4CLIP · MSR-VTT 1k-A test | Text-to-video R@1: 44.9 -> 45.3 | Improved |
| Use salient sampling with 2 selected frames instead of sparse sampling (ablation) | LGDN · DiDeMo test | R@SUM: 183.0 -> 198.3 | Improved |
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.
- TACo — TACo: Token-Aware Cascade Contrastive Learning for Video-Text Alignment (ICCV 2021) · better on 1 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
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.88)
- LanguageBind: Extending Video-Language Pretraining to N-modality by Language-based Semantic Alignment (ICLR 2024 · similarity 0.87)
- VidLA: Video-Language Alignment at Scale (CVPR 2024 · similarity 0.87)
- Video-LaVIT: Unified Video-Language Pre-training with Decoupled Visual-Motional Tokenization (ICML 2024 · similarity 0.87)
- TOPA: Extending Large Language Models for Video Understanding via Text-Only Pre-Alignment (NeurIPS 2024 · similarity 0.87)
- Video-Panda: Parameter-efficient Alignment for Encoder-free Video-Language Models (CVPR 2025 · similarity 0.86)
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.
