Research map / Video / Video-language understanding
Flow Graph to Video Grounding for Weakly-Supervised Multi-step Localization
Nikita Dvornik, Isma Hadji, Hai Pham, Dhaivat Bhatt, Brais Martinez, Afsaneh Fazly, Allan D. Jepson
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 |
|---|---|---|---|
| ground the flow graph with Graph2Vid instead of aligning a linear procedure | step localization at inference · CrossTask | Acc.: 22.3 -> 25.3 | Improved |
| ground the flow graph with Graph2Vid instead of aligning a linear procedure | step localization at inference · CrossTask | IoU: 15.1 -> 17.1 | Improved |
| use linear procedure order instead of unordered bag of steps | step localization at inference · CrossTask | Acc.: 20.5 -> 22.3 | Improved |
| train with Graph2Vid loss instead of Drop-DTW on linear procedure | MLP on pre-trained features · CrossTask | Acc.: 25.0 -> 26.3 | Improved |
| train with Graph2Vid loss instead of soft clustering on bag of steps | MLP on pre-trained features · CrossTask | IoU: 17.3 -> 19.1 | Improved |
| train with linear-procedure Drop-DTW versus no training | MLP on pre-trained features · CrossTask | Acc.: 25.3 -> 25.0 | Hurt |
| train with Graph2Vid versus no training | MLP on pre-trained features · CrossTask | IoU: 17.1 -> 19.1 | Improved |
| use learning-based parser graphs instead of rule-based parser graphs (ablation) | Graph2Vid · CrossTask | Acc.: 24.3 -> 25.3 | Improved |
| use learning-based parser graphs instead of manual annotation (ablation) | Graph2Vid · CrossTask | Acc.: 24.8 -> 25.3 | Improved |
| use manual graphs instead of rule-based parser graphs (ablation) | Graph2Vid · CrossTask | IoU: 16.3 -> 16.8 | Improved |
| use tSort graph algorithm instead of brute-force enumeration of topological sorts | flow graph grounding · CrossTask | grounding time: 3.2 * 10^4 ms -> 57 ms | Improved |
| use ground-truth step order instead of flow graph | step localization at inference · CrossTask | Acc.: 25.3 -> 32.4 | Improved |
Closest papers by meaning
- StepFormer: Self-Supervised Step Discovery and Localization in Instructional Videos (CVPR 2023 · similarity 0.91)
- Weakly Supervised Temporal Action Localization via Representative Snippet Knowledge Propagation (CVPR 2022 · similarity 0.88)
- HR-Pro: Point-Supervised Temporal Action Localization via Hierarchical Reliability Propagation (AAAI 2024 · similarity 0.87)
- What When and Where? Self-Supervised Spatio-Temporal Grounding in Untrimmed Multi-Action Videos from Narrated Instructions (CVPR 2024 · similarity 0.86)
- Omnipotent Distillation with LLMs for Weakly-Supervised Natural Language Video Localization: When Divergence Meets Consistency (AAAI 2024 · similarity 0.86)
- VideoGEM: Training-free Action Grounding in Videos (CVPR 2025 · similarity 0.86)
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.
