Research map / Video / Action recognition
Encouraging LSTMs to Anticipate Actions Very Early
Mohammad Sadegh Aliakbarian, Fatemeh Sadat Saleh, Mathieu Salzmann, Basura Fernando, Lars Petersson, Lars Andersson
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 action-aware features to context-aware features via multi-stage LSTM (ablation) | MS-LSTM with anticipation loss · UCF-101 anticipation | accuracy, earliest: 62.80 -> 80.5 | Improved |
| add context-aware features to action-aware features via multi-stage LSTM (ablation) | MS-LSTM with anticipation loss · UCF-101 anticipation | accuracy, earliest: 69.33 -> 80.5 | Improved |
| add context-aware features to action-aware features via multi-stage LSTM (ablation) | MS-LSTM with anticipation loss · UCF-101 anticipation | accuracy, latest: 77.86 -> 83.37 | Improved |
| process features sequentially (multi-stage) instead of concatenating (ablation) | Ours MS-LSTM · UCF-101 recognition (split 1) | accuracy: 77.16 -> 83.37 | Improved |
| process features sequentially instead of two parallel LSTMs (ablation) | Ours MS-LSTM · UCF-101 recognition (split 1) | accuracy: 78.63 -> 83.37 | Improved |
| process context-aware features first, then action-aware (not swapped) (ablation) | Ours MS-LSTM · UCF-101 recognition (split 1) | accuracy: 78.80 -> 83.37 | Improved |
| add the LSTM stages on top of the CNN (ablation) | Ours · UCF-101 anticipation | accuracy, earliest: 69.53 -> 80.5 | Improved |
| add the LSTM stages on top of the CNN (ablation) | Ours · UCF-101 anticipation | accuracy, latest: 73.80 -> 83.4 | Improved |
| add average pooling of predictions over time (ablation) | Ours · UCF-101 recognition (split 1) | accuracy: 82.16 -> 83.37 | Improved |
| increase hidden units from 1024 to 2048 (ablation) | Ours · UCF-101 recognition (split 1) | accuracy: 81.94 -> 82.16 | Improved |
| replace cross-entropy loss with the proposed anticipation loss (ablation) | MS-LSTM · UCF-101 recognition (split 1) | accuracy: 78.93 -> 83.37 | Improved |
| replace linearly growing loss with the proposed anticipation loss (ablation) | MS-LSTM · UCF-101 recognition (split 1) | accuracy: 81.27 -> 83.37 | Improved |
Closest papers by meaning
- Rethinking Learning Approaches for Long-Term Action Anticipation (ECCV 2022 · similarity 0.91)
- Time-Conditioned Action Anticipation in One Shot (CVPR 2019 · similarity 0.90)
- Rolling-Unrolling LSTMs for Action Anticipation from First-Person Video (TPAMI 2020 · similarity 0.90)
- Can't Make an Omelette Without Breaking Some Eggs: Plausible Action Anticipation Using Large Video-Language Models (CVPR 2024 · similarity 0.87)
- AntGPT: Can Large Language Models Help Long-term Action Anticipation from Videos? (ICLR 2024 · similarity 0.87)
- ActFusion: a Unified Diffusion Model for Action Segmentation and Anticipation (NeurIPS 2024 · similarity 0.87)
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.
