Research map / Video / Action recognition
Rolling-Unrolling LSTMs for Action Anticipation from First-Person Video
Antonino Furnari, Giovanni Maria Farinella
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 the rolling-unrolling two-LSTM mechanism to the single-LSTM baseline (late fusion) (ablation) | BL (Late) · EPIC-Kitchens val ablation (anticipation 1.0s) | Top-5 action accuracy (%): 32.02 -> 34.23 | Improved |
| replace late fusion with Modality ATTention (MATT) (ablation) | RU · EPIC-Kitchens val ablation (anticipation 1.0s) | Top-5 action accuracy (%): 34.23 -> 35.32 | Improved |
| replace early fusion with Modality ATTention (MATT) (ablation) | RU · EPIC-Kitchens val ablation (anticipation 1.0s) | Top-5 action accuracy (%): 31.88 -> 35.32 | Improved |
| add Sequence Completion Pre-Training (SCP) (ablation) | RU · EPIC-Kitchens val ablation (anticipation 1.0s) | Top-5 action accuracy (%): 34.75 -> 35.32 | Improved |
| add OBJ modality to RGB+Flow (ablation) | RU with MATT · EPIC-Kitchens val ablation (anticipation 1.0s) | Top-5 action accuracy (%): 32.16 -> 35.32 | Improved |
| add optical flow to RGB+OBJ (ablation) | RU with MATT · EPIC-Kitchens val ablation (anticipation 1.0s) | Top-5 action accuracy (%): 34.27 -> 35.32 | Improved |
| add RGB to Flow+OBJ (ablation) | RU with MATT · EPIC-Kitchens val ablation (anticipation 1.0s) | Top-5 action accuracy (%): 23.93 -> 35.32 | Improved |
| fuse all three modalities versus the RGB branch alone (ablation) | RU · EPIC-Kitchens val ablation (anticipation 1.0s) | Top-5 action accuracy (%): 30.83 -> 35.32 | Improved |
| add rolling-unrolling, MATT and SCP together to the strong baseline (ablation) | BL · EPIC-Kitchens val ablation (anticipation 1.0s) | Top-5 action accuracy (%): 32.02 -> 35.32 | Improved |
| vary the number of encoding steps S_enc (ablation) | RU · EPIC-Kitchens val | Top-5 action accuracy (%) | Mixed |
| add unsupervised future-representation pre-training to ED (ablation) | ED · EPIC-Kitchens val (anticipation 1.0s) | Top-5 action accuracy (%): 25.20 -> 25.75 | No clear effect |
| add unsupervised future-representation pre-training to ED (ablation) | ED · ActivityNet (anticipation 1.0s) | Top-5 action accuracy (%): 65.84 -> 72.93 | Improved |
Closest papers by meaning
- What Would You Expect? Anticipating Egocentric Actions With Rolling-Unrolling LSTMs and Modality Attention (ICCV 2019 · similarity 0.94)
- Encouraging LSTMs to Anticipate Actions Very Early (ICCV 2017 · similarity 0.90)
- LSTA: Long Short-Term Attention for Egocentric Action Recognition (CVPR 2019 · similarity 0.89)
- Summarize the Past to Predict the Future: Natural Language Descriptions of Context Boost Multimodal Object Interaction Anticipation (CVPR 2024 · similarity 0.87)
- Action Scene Graphs for Long-Form Understanding of Egocentric Videos (CVPR 2024 · similarity 0.86)
- Test-Time Adaptation for Combating Missing Modalities in Egocentric Videos (ICLR 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.
