Research map / Video / Video segmentation and tracking
How Local is the Local Diversity? Reinforcing Sequential Determinantal Point Processes with Dynamic Ground Sets for Supervised Video Summarization
Yandong Li, Liqiang Wang, Tianbao Yang, Boqing Gong
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 |
|---|---|---|---|
| sample trajectories from the policy instead of greedily (ablation) | DySeqDPP · UTE (K=12) | F1-score: 31.09 -> 33.45 | Improved |
| weight only the final whole-summary reward (gamma=1e-20) instead of gamma=0.9 with full-summary reward (ablation) | DySeqDPP · UTE (K=12) | F1-score: 32.16 -> 33.45 | Improved |
| compare incomplete summary with user summary up to current step instead of full summary (gamma=0.9) (ablation) | DySeqDPP · UTE (K=12) | F1-score: 32.16 -> 32.77 | Improved |
| compare incomplete summary with user summary up to current step instead of full summary (gamma=0.2) (ablation) | DySeqDPP · UTE (K=12) | F1-score: 33.08 -> 33.15 | No clear effect |
| use partial instead of full reward at gamma=0.5 (ablation) | DySeqDPP · UTE (K=12) | F1-score: 32.96 -> 33.01 | No clear effect |
| pool whole video to predict partition instead of current segment only (PoolSeg) (ablation) | DySeqDPP · UTE (K=12) | F1-score: 31.33 -> 32.71 | Improved |
| use summary and current-segment features instead of pooling whole video (PoolVideo) (ablation) | DySeqDPP · UTE (K=12) | F1-score: 32.71 -> 33.45 | Improved |
| learn dynamic partition and train with reinforcement learning instead of SeqDPP | SeqDPP · SumMe | F1-score: 40.8 ± 4.8 -> 44.3 ± 2.8 | Improved |
| learn dynamic partition and train with reinforcement learning instead of SeqDPP | SeqDPP · TVSum | F1-score: 57.4 ± 2.0 -> 58.4 ± 2.5 | Improved |
| change SeqDPP segment size among 5, 10 and 12 shots | SeqDPP · UTE | F1-score | No clear effect |
| train SeqDPP with the reinforcement learning algorithm instead of original training | SeqDPP · UTE | F1-score | Improved |
| add a latent variable for dynamic partitioning of segments instead of fixed equal segments | SeqDPP with RL · UTE | F1-score | Improved |
Closest papers by meaning
- Deep Reinforcement Learning for Unsupervised Video Summarization With Diversity-Representativeness Reward (AAAI 2018 · similarity 0.86)
- Video Summarization by Learning From Unpaired Data (CVPR 2019 · similarity 0.85)
- FastVID: Dynamic Density Pruning for Fast Video Large Language Models (NeurIPS 2025 · similarity 0.84)
- Video Summarization Using Denoising Diffusion Probabilistic Model (AAAI 2025 · similarity 0.84)
- FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 2024 · similarity 0.84)
- SyncVIS: Synchronized Video Instance Segmentation (NeurIPS 2024 · 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.
