Research map / Video / Video-language understanding
SAIL: Similarity-Aware Guidance and Inter-Caption Augmentation-based Learning for Weakly-Supervised Dense Video Captioning
Ye-Chan Kim, SeungJu Cha, Si-Woo Kim, Minju Jeon, Hyungee Kim, Dong-Jin Kim
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 similarity-aware mask guide (Lsim) (ablation) | SAIL / ILCACM · ActivityNet Captions val | CIDEr: 33.42 -> 35.18 | Improved |
| add LLM synthetic captions with inter-masks (ablation) | SAIL / ILCACM · ActivityNet Captions val | CIDEr: 33.42 -> 34.92 | Improved |
| combine similarity mask and synthetic captions (ablation) | SAIL · ActivityNet Captions val | CIDEr: 35.18 -> 35.38 | Improved |
| use synthetic captions as hard negatives in Lsim rather than as auxiliary inter-mask signal (ablation) | SAIL · ActivityNet Captions val | CIDEr: 35.38 -> 33.84 | Hurt |
| increase synthetic caption ratio from 25% to 100% (ablation) | SAIL · ActivityNet Captions val | CIDEr: 33.76 -> 35.38 | Improved |
| use Gaussian masks instead of Cauchy masks (ablation) | SAIL · ActivityNet Captions val | CIDEr: 33.85 -> 35.38 | Improved |
| use Gaussian masks instead of hard binary masks (ablation) | SAIL · ActivityNet Captions val | CIDEr: 17.54 -> 35.38 | Improved |
| apply SAIL on top of ILCACM with Hard Binary mask (ablation) | ILCACM · ActivityNet Captions val (Hard Binary mask) | CIDEr: 16.96 -> 17.54 | Improved |
| vary inter-mask width w_inter (ablation) | SAIL · ActivityNet Captions val | CIDEr, F1 | No clear effect |
| use SAIL instead of ILCACM on computation | SAIL · ActivityNet Captions val | training time: 1H 38M 27S -> 1H 41M 59S | No clear effect |
| train SAIL instead of ILCACM on YouCook2 | SAIL · YouCook2 val | CIDEr: 13.49 -> 14.61 | Improved |
| reduce ground-truth caption density to 25% (baseline) | ILCACM · ActivityNet Captions val | CIDEr, SODA_c | Hurt |
Closest papers by meaning
- Implicit Location-Caption Alignment via Complementary Masking for Weakly-Supervised Dense Video Captioning (AAAI 2025 · similarity 0.91)
- Explicit Temporal-Semantic Modeling for Dense Video Captioning via Context-Aware Cross-Modal Interaction (AAAI 2026 · similarity 0.88)
- Jointly Localizing and Describing Events for Dense Video Captioning (CVPR 2018 · similarity 0.87)
- Do You Remember? Dense Video Captioning with Cross-Modal Memory Retrieval (CVPR 2024 · similarity 0.85)
- Dense Video Object Captioning from Disjoint Supervision (ICLR 2025 · similarity 0.85)
- VidChain: Chain-of-Tasks with Metric-based Direct Preference Optimization for Dense Video Captioning (AAAI 2025 · similarity 0.85)
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.
