Research map / Video / Action recognition
Ego4D Goal-Step: Toward Hierarchical Understanding of Procedural Activities
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| train jointly on steps and substeps instead of steps only (ablation) | VSLNet · step grounding, step queries | Recall@1, mIoU=0.3: 13.9 -> 15.8 | Helped |
| train jointly on steps and substeps instead of substeps only (ablation) | VSLNet · step grounding, substep queries | Recall@1, mIoU=0.3: 6.4 -> 8.7 | Helped |
| train jointly on steps and substeps instead of substeps only (ablation) | VSLNet · step grounding, step queries | Recall@1, mIoU=0.3: 9.6 -> 15.8 | Helped |
| pair substeps with parent steps descriptions instead of joint training (ablation) | VSLNet · step grounding, step queries | Recall@1, mIoU=0.3: 15.8 -> 7.5 | Hurt |
| replace steps/substeps with equal amount of narrations (ablation) | VSLNet · step grounding, substep queries | Recall@1, mIoU=0.3: 8.7 -> 7.4 | Hurt |
| add narrations to steps instead of substeps (ablation) | VSLNet · step grounding, step queries | Recall@1, mIoU=0.3: 13.9 -> 14.0 | No clear effect |
| combine goal and step/substep training instances (ablation) | ActionFormer · goal localization | detection mAP: 42.2 -> 37.1 | Hurt |
| combine goal and step/substep training instances (ablation) | ActionFormer · step localization | detection mAP: 9.9 -> 9.7 | No clear effect |
| combine goal and step/substep training instances (ablation) | LSTR · online goal detection | per-frame mAP: 21.5 -> 21.2 | No clear effect |
| combine goal and step/substep training instances (ablation) | LSTR · online step detection | per-frame mAP: 8.7 -> 8.1 | Hurt |
| finetune ViT directly on dataset (EgoOnly) instead of Omnivore features | ActionFormer · step localization val | detection mAP: 9.9 -> 13.1 | Helped |
| finetune ViT directly on dataset (EgoOnly) instead of Omnivore features | LSTR · online step detection val | per-frame mAP: 8.7 -> 10.2 | Helped |
Closest papers by meaning
- CaptainCook4D: A Dataset for Understanding Errors in Procedural Activities (NeurIPS 2024 · similarity 0.87)
- EgoExoLearn: A Dataset for Bridging Asynchronous Ego- and Exo-centric View of Procedural Activities in Real World (CVPR 2024 · similarity 0.87)
- Action Scene Graphs for Long-Form Understanding of Egocentric Videos (CVPR 2024 · similarity 0.87)
- Differentiable Task Graph Learning: Procedural Activity Representation and Online Mistake Detection from Egocentric Videos (NeurIPS 2024 · similarity 0.87)
- Ego-Exo4D: Understanding Skilled Human Activity from First- and Third-Person Perspectives (CVPR 2024 · similarity 0.87)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · 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.