Research map / Video / Video-language understanding
SHINE: Saliency-aware HIerarchical NEgative Ranking for Compositional Temporal Grounding
Zixu Cheng, Yujiang Pu, Shaogang Gong, Parisa Kordjamshidi, Yu Kong
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 SHINE (LLM hard negatives + coarse-to-fine saliency ranking) to QD-DETR | QD-DETR · Charades-CG Novel-Composition | R1@0.5: 42.30 -> 50.23 | Improved |
| add SHINE to Moment-DETR | Moment-DETR · Charades-CG Test-Trivial | R1@0.5: 49.48 -> 57.14 | Improved |
| add SHINE to QD-DETR on ActivityNet-CG | QD-DETR · ActivityNet-CG Novel-Composition | R1@0.5: 26.91 -> 29.56 | Improved |
| add SHINE to QD-DETR on Novel-Word split of ActivityNet-CG | QD-DETR · ActivityNet-CG Novel-Word | mIoU: 31.21 -> 30.98 | No clear effect |
| add L_inter (positive vs negative query saliency gap) to coarse-grained loss (ablation) | QD-DETR · Charades-CG Novel-Composition | R1@0.5: 42.30 -> 46.69 | Improved |
| replace base saliency loss with L_intra (ablation) | QD-DETR · Charades-CG Novel-Composition | R1@0.5: 42.30 -> 44.02 | Improved |
| add first fine-grained ranking constraint L_fr^1 (ablation) | QD-DETR · Charades-CG Novel-Composition | R1@0.5: 42.30 -> 46.43 | Improved |
| add L_fr^3 without the complete hierarchy (ablation) | QD-DETR · Charades-CG Novel-Composition | R1@0.5: 46.72 -> 46.40 | Hurt |
| complete all four fine-grained constraints together with L_cr (ablation) | QD-DETR · Charades-CG Novel-Composition | R1@0.7: 24.52 -> 27.69 | Improved |
| use LLM (GPT-3.5 Turbo) hard negatives instead of random sampling (ablation) | QD-DETR · Charades-CG Novel-Composition | R1@0.5: 47.41 -> 50.23 | Improved |
| use LLM hard negatives instead of random sampling on ActivityNet-CG (ablation) | QD-DETR · ActivityNet-CG Novel-Composition | R1@0.5: 29.59 -> 29.56 | No clear effect |
| include prepositions and adverbs in hard negative construction (ablation) | QD-DETR · Charades-CG Novel-Composition | R1@0.7: 25.28 -> 27.69 | 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.
- Deco — DeCo: Decomposition and Reconstruction for Compositional Temporal Grounding via Coarse-To-Fine Contrastive Ranking (CVPR 2023) · 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
- Compositional Temporal Grounding With Structured Variational Cross-Graph Correspondence Learning (CVPR 2022 · similarity 0.91)
- Negative Sample Matters: A Renaissance of Metric Learning for Temporal Grounding (AAAI 2022 · similarity 0.90)
- PC-Net: Weakly Supervised Compositional Moment Retrieval via Proposal-Centric Network (NeurIPS 2025 · similarity 0.90)
- Universal Video Temporal Grounding with Generative Multi-modal Large Language Models (NeurIPS 2025 · similarity 0.89)
- Empower Words: DualGround for Structured Phrase and Sentence-Level Temporal Grounding (NeurIPS 2025 · similarity 0.89)
- Generative Video Diffusion for Unseen Novel Semantic Video Moment Retrieval (AAAI 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.
