Research map / Label-efficient and robust learning / Domain adaptation and generalization
DDLSTM: Dual-Domain LSTM for Cross-Dataset Action Recognition
Toby Perrett, Dima Damen
Check 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 |
|---|---|---|---|
| Replace standard batch normalization with learned dual-domain cross-contamination in input-to-hidden and hidden-to-hidden LSTM weights | LSTM trained jointly on two datasets · 50 Salads, MPII Cooking 2, and Breakfast dataset pairs | frame-level accuracy: 5.1%; 2.5%; 3.7%; 3.8%; 4.2%; 1.0% | Improved |
| Use jointly trained DDLSTM rather than jointly trained LSTM | 50 Salads (paired with MPII Cooking 2) · 50 Salads | frame-level accuracy: 41.6 -> 47.1 | Improved |
| Use jointly trained DDLSTM rather than jointly trained LSTM | MPII Cooking 2 (paired with 50 Salads) · MPII Cooking 2 | frame-level accuracy: 39.0 -> 41.5 | Improved |
| Use jointly trained DDLSTM rather than jointly trained LSTM | Breakfast (paired with 50 Salads) · Breakfast | frame-level accuracy: 24.7 -> 29.1 | Improved |
| Use jointly trained DDLSTM rather than pre-training/fine-tuning LSTM | 50 Salads (paired with Breakfast) · 50 Salads | frame-level accuracy: 42.5 -> 46.3 | Improved |
| Use jointly trained DDLSTM rather than jointly trained LSTM | Breakfast (paired with MPII Cooking 2) · Breakfast | frame-level accuracy: 26.3 -> 30.5 | Improved |
| Use jointly trained DDLSTM rather than jointly trained LSTM | MPII Cooking 2 (paired with Breakfast) · MPII Cooking 2 | frame-level accuracy: 39.1 -> 40.1 | Improved |
| Use 3 LSTM layers rather than 1 (ablation) | DDLSTM · 50 Salads, MPII Cooking 2, and Breakfast dataset pairs | frame-level accuracy: 45.7 -> 47.1; 41.3 -> 41.5; 28.4 -> 29.1; 45.0 -> 46.3; 29.2 -> 30.5; 40.8 -> 40.1 | Mixed |
| Use 3 LSTM layers rather than 2 (ablation) | DDLSTM · 50 Salads, MPII Cooking 2, and Breakfast dataset pairs | frame-level accuracy: 46.6 -> 47.1; 41.0 -> 41.5; 29.0 -> 29.1; 45.2 -> 46.3; 29.9 -> 30.5; 40.1 -> 40.1 | Mixed |
| Use 3 LSTM layers rather than 4 (ablation) | DDLSTM · 50 Salads, MPII Cooking 2, and Breakfast dataset pairs | frame-level accuracy: 46.1 -> 47.1; 40.6 -> 41.5; 28.6 -> 29.1; 44.3 -> 46.3; 30.5 -> 30.5; 38.8 -> 40.1 | Mixed |
| Use 3 LSTM layers rather than 5 (ablation) | DDLSTM · 50 Salads, MPII Cooking 2, and Breakfast dataset pairs | frame-level accuracy: 41.0 -> 47.1; 40.3 -> 41.5; 29.0 -> 29.1; 46.0 -> 46.3; 30.3 -> 30.5; 38.4 -> 40.1 | Improved |
| Use 3 LSTM layers rather than 10 (ablation) | DDLSTM · 50 Salads, MPII Cooking 2, and Breakfast dataset pairs | frame-level accuracy: 39.7 -> 47.1; 38.0 -> 41.5; 25.5 -> 29.1; 40.7 -> 46.3; 28.1 -> 30.5; 35.6 -> 40.1 | Improved |
Closest papers by meaning
- Beyond Sharing Weights for Deep Domain Adaptation (TPAMI 2018 · similarity 0.86)
- MultiDIAL: Domain Alignment Layers for (Multisource) Unsupervised Domain Adaptation (TPAMI 2020 · similarity 0.86)
- Interact Before Align: Leveraging Cross-Modal Knowledge for Domain Adaptive Action Recognition (CVPR 2022 · similarity 0.86)
- Multi-Modal Domain Adaptation for Fine-Grained Action Recognition (CVPR 2020 · similarity 0.86)
- Self-Paced Collaborative and Adversarial Network for Unsupervised Domain Adaptation (TPAMI 2019 · similarity 0.85)
- Adversarial Cross-Domain Action Recognition with Co-Attention (AAAI 2020 · 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.
