Research map / Video / Video segmentation and tracking
TRINITY: A Multi-Perspective Benchmark for Personal-Style Video Highlight Detection
Qianqian Chen, Hyun Bin Kim, Denzel Wijaya, Yang Yi, Bo LIU, Yangkai Ding
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 shared-backbone multi-task learning over single-task training (ablation) | Single-task baseline · TRINITY (avg) | mAP rho=15%: +3.50 | Improved |
| add shared and private experts instead of independent experts (ablation) | Independent experts (S3) · TRINITY-Nature | mAP rho=15%: 44.78 to 58.74 | Improved |
| add shared and private experts instead of independent experts (ablation) | Independent experts (S3) · TRINITY-Event | mAP rho=15%: 37.54 to 40.98 | Improved |
| add shared and private experts instead of independent experts (ablation) | Independent experts (S3) · TRINITY-Emotion | mAP rho=15%: 63.27 to 67.42 | Improved |
| use independent experts instead of shared-backbone MTL (ablation) | Shared-backbone MTL (S2) · TRINITY (avg) | mAP rho=50%: 68.97 -> 69.70 | Improved |
| add Nature supervision to Event-only training (ablation) | Shared-private model · TRINITY-Nature | mAP rho=15%: 31.25 to 57.89 | Improved |
| add Nature supervision to Event-only training (ablation) | Shared-private model · TRINITY-Event | mAP rho=15%: 40.25 -> 39.98 | No clear effect |
| add Emotion supervision to Event+Nature training (ablation) | Shared-private model · TRINITY-Emotion | mAP rho=15%: 45.98 -> 67.42 | Improved |
| use MixBatch perspective mixing instead of per-task batches (ablation) | Ours · TRINITY-Nature | mAP rho=15%: 58.74 to 42.19 | Hurt |
| use MixBatch perspective mixing instead of per-task batches (ablation) | Ours · TRINITY-Emotion | mAP rho=15%: 67.42 -> 63.61 | Mixed |
| add round-robin training instead of sequential for single backbone (ablation) | Single backbone · TRINITY (avg) | mAP rho=15%: 48.89 -> 49.19 | Improved |
| add round-robin training instead of sequential (ablation) | Shared-private model · TRINITY (avg) | mAP rho=15%: 55.56 -> 55.71 | 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.
- SummDiff — SummDiff: Generative Modeling of Video Summarization with Diffusion (ICCV 2025) · better on 2 of 2 shared results
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
- Learning Pixel-Level Distinctions for Video Highlight Detection (CVPR 2022 · similarity 0.87)
- MINI-Net: Multiple Instance Ranking Network for Video Highlight Detection (ECCV 2020 · similarity 0.87)
- HighlightMe: Detecting Highlights From Human-Centric Videos (ICCV 2021 · similarity 0.87)
- Bridging the Gap: A Unified Video Comprehension Framework for Moment Retrieval and Highlight Detection (CVPR 2024 · similarity 0.85)
- Exploring Domain Incremental Video Highlights Detection with the LiveFood Benchmark (AAAI 2024 · similarity 0.84)
- ShotVL: Human-Centric Highlight Frame Retrieval via Language Queries (AAAI 2025 · similarity 0.84)
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.
