Research map / Video / Video-language understanding
Towards Diverse Paragraph Captioning for Untrimmed Videos
Yuqing Song, Shizhe Chen, Qin Jin
Check 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 |
|---|---|---|---|
| Replace event-proposal captioning with direct paragraph generation | Video paragraph captioning · ActivityNet Captions ae-test | CIDEr: 24.14 -> 27.36 | Improved |
| Add progressive memory exposure (ablation) | Vanilla video paragraph captioner · ActivityNet ae-test | Div@1: 64.92 -> 66.79 | Improved |
| Add over-accessed memory decay (ablation) | Vanilla video paragraph captioner · ActivityNet ae-test | Div@1: 64.92 -> 66.18 | Improved |
| Add both progressive memory exposure and over-accessed memory decay (ablation) | Vanilla video paragraph captioner · ActivityNet ae-test | CIDEr: 24.11 -> 25.65 | Improved |
| Add token-level high-frequency penalty (ablation) | Captioner with both dynamic memory mechanisms · ActivityNet ae-test | Div@1: 67.90 -> 68.42 | Improved |
| Add CIDEr relevance reward in RL (ablation) | Captioner with dynamic memories and token penalty · ActivityNet ae-test | CIDEr: 26.55 -> 27.06 | Improved |
| Add phrase-level high-frequency penalty to the CIDEr relevance reward (ablation) | Captioner with dynamic memories, token penalty, and relevance reward · ActivityNet ae-test | Div@1: 67.81 -> 68.33 | Improved |
| Add CIDEr relevance reward without phrase-level penalty (ablation) | Captioner with dynamic memories and token penalty · ActivityNet ae-test | Div@1: 68.42 -> 67.81 | Hurt |
| Add video reconstruction loss for keyframe selection (ablation) | MLE-trained keyframe-aware captioner · ActivityNet ae-test, keyframe ratio 0.5 | CIDEr: 25.47 -> 26.80 | Improved |
| Add sparsity loss for keyframe selection (ablation) | MLE-trained keyframe-aware captioner · ActivityNet ae-test, keyframe ratio 0.5 | CIDEr: 25.47 -> 26.02 | Improved |
| Use learned keyframe selection instead of uniform interval sampling | Keyframe-aware video encoder · ActivityNet ae-test | CIDEr | Improved |
| Reduce the keyframe selection ratio | Keyframe-aware video encoder · ActivityNet ae-test | CIDEr | Mixed |
Closest papers by meaning
- Unifying Event Detection and Captioning as Sequence Generation via Pre-training (ECCV 2022 · similarity 0.91)
- Event-Equalized Dense Video Captioning (CVPR 2025 · similarity 0.90)
- Sketch, Ground, and Refine: Top-Down Dense Video Captioning (CVPR 2021 · similarity 0.90)
- Move Forward and Tell: A Progressive Generator of Video Descriptions (ECCV 2018 · similarity 0.90)
- DIBS: Enhancing Dense Video Captioning with Unlabeled Videos via Pseudo Boundary Enrichment and Online Refinement (CVPR 2024 · similarity 0.87)
- Set Prediction Guided by Semantic Concepts for Diverse Video Captioning (AAAI 2024 · 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.
