Research map / Video / Video-language understanding
Fine-grained Dynamic Network for Generic Event Boundary Detection
Ziwei Zheng, Lijun He, Le Yang, Fan Li
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 pairwise contrast module (PCM) to the detector (ablation) | DyBDet detector · Kinetics-GEBD val | F1@0.05: 52.6 -> 76.3 | Improved |
| add multi-order difference encoder (MDE) on top of PCM (ablation) | DyBDet detector · Kinetics-GEBD val | F1@0.05: 76.3 -> 78.5 | Improved |
| add cross-SE fusion of PCM and MDE outputs (ablation) | DyBDet detector · Kinetics-GEBD val | F1@0.05: 78.5 -> 79.6 | Improved |
| use multi-order temporal differences instead of heavier mixers (self-attention, Conv1d, more blocks) (ablation) | MDE · Kinetics-GEBD val | F1@0.05: 78.5 -> 79.6 | Improved |
| use different difference orders per detector (0; +1; +1+2) instead of the same orders in all (ablation) | DyBDet detectors · Kinetics-GEBD val | F1@0.05: 78.9 -> 79.6 | Improved |
| learn weights over difference orders instead of fixed per-detector choice (ablation) | DyBDet detectors · Kinetics-GEBD val | F1@0.05: 79.6 -> 78.2 | Hurt |
| use partial exit criterion instead of post-combining detector outputs (ablation) | DyBDet · Kinetics-GEBD val | F1@0.05: 78.1 -> 79.6 | Improved |
| pad exited positions with repeated nearby features rather than linear interpolation or none (ablation) | partial exit · Kinetics-GEBD val | F1@0.05: 79.0 -> 79.6 | Improved |
| increase local window size beyond 17 (ablation) | DyBDet · Kinetics-GEBD val | F1@0.05: 79.6 -> 79.5 | No clear effect |
| swap in the heavy SC-Transformer detector within the dynamic network (ablation) | dynamic multi-exit network · Kinetics-GEBD val | F1@0.05: -1.9% | Hurt |
| use PCM-only detector in the dynamic network (ablation) | dynamic multi-exit network · Kinetics-GEBD val | F1@0.05 vs compute | No clear effect |
| replace CSN backbone for ResNet50 (stronger backbone) | DyBDet · Kinetics-GEBD val | F1@0.05: 79.6 -> 83.1 | 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.
- Temporal Perceiver — Temporal Perceiver: A General Architecture for Arbitrary Boundary Detection (TPAMI 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
- Online Generic Event Boundary Detection (ICCV 2025 · similarity 0.94)
- Generic Event Boundary Detection: A Benchmark for Event Segmentation (ICCV 2021 · similarity 0.92)
- Generic Event Boundary Detection via Denoising Diffusion (ICCV 2025 · similarity 0.92)
- UBoCo: Unsupervised Boundary Contrastive Learning for Generic Event Boundary Detection (CVPR 2022 · similarity 0.90)
- VADTree: Explainable Training-Free Video Anomaly Detection via Hierarchical Granularity-Aware Tree (NeurIPS 2025 · similarity 0.85)
- DiffDVC: Accurate Event Detection for Dense Video Captioning via Diffusion Models (AAAI 2025 · similarity 0.85)
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.
