Research map / Video / Video-language understanding
Bayesian Modeling of Temporal Coherence in Videos for Entity Discovery and Summarization
Adway Mitra, Soma Biswas, Chiranjib Bhattacharyya
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 joint temporal-segment modeling (TC-CRF over TC-CRP) | TC-CRP · Maha65 entity discovery | Purity: 0.89 -> 0.97 | Improved |
| add temporal-coherence prior (TC-CRP over sHDP-HMM) | sHDP-HMM · BBTs1e3 entity discovery | Purity: 0.76 -> 0.83 | Improved |
| use TC-CRF instead of TC-CRP | TC-CRP · Maha66 entity discovery | Purity: 0.96 -> 0.91 | Hurt |
| Bayesian nonparametric methods vs constrained spectral clustering | TC-CRF · BBTs1e3 entity discovery | Tracklet coverage: 0.09 -> 0.86 | Improved |
| add TC-CRF over sHDP-HMM | sHDP-HMM · Maha81 entity discovery | Tracklet coverage: 0.66 -> 0.80 | Improved |
| reduce tracklet size from R=10 to R=5 or R=1 (ablation) | TC-CRF, TC-CRP, sHDP-HMM · entity discovery | Purity/coverage | No clear effect |
| reduce tracklet size to R=1 (ablation) | SBMR · entity discovery | matrix rank: 120-130 | Hurt |
| online single-pass inference instead of offline Gibbs sampling | TC-CRF · Maha81 entity discovery | Purity: 0.89 -> 0.71 | Hurt |
| TC-CRP junk component variance factor c from 2 to 4 (ablation) | TC-CRP · Maha65 non-face discovery | Precision: 0.88 -> 0.98 | Improved |
| TC-CRP junk component variance factor c from 2 to 5 (ablation) | TC-CRP · Maha65 non-face discovery | Recall*: 106 -> 79 | Hurt |
| TC-CRP junk detection instead of KMeans | KMeans · Maha65 non-face discovery | Precision: 0.22 -> 0.98 | Improved |
| add TC-CRF over TC-CRP for tracklet linking | TC-CRP · Maha82 tracklet linking | Fraction fully linked: 0.86 -> 0.89 | Improved |
Closest papers by meaning
- Hierarchical Long Video Understanding with Audiovisual Entity Cohesion and Agentic Search (CVPR 2026 · similarity 0.84)
- Reading the Videos: Temporal Labeling for Crowdsourced Time-Sync Videos Based on Semantic Embedding (AAAI 2016 · similarity 0.83)
- Query-Focused Extractive Video Summarization (ECCV 2016 · similarity 0.83)
- Query-Focused Video Summarization: Dataset, Evaluation, and a Memory Network Based Approach (CVPR 2017 · similarity 0.83)
- EntitySAM: Segment Everything in Video (CVPR 2025 · similarity 0.82)
- EntityBench: Towards Entity-Consistent Long-Range Multi-Shot Video Generation (NeurIPS 2026 · similarity 0.82)
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.
