Research map / Label-efficient and robust learning / Domain adaptation and generalization
CycDA: Unsupervised Cycle Domain Adaptation to Learn from Image to Video
Wei Lin, Anna Kukleva, Kunyang Sun, Horst Possegger, Hilde Kuehne, Horst Bischof
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 video model trained on image-model pseudo labels (ablation) | CycDA · EADs->HMDB51 split 1 | accuracy: 39.0 -> 50.5 | Improved |
| add class-agnostic domain alignment in stage 1 (ablation) | CycDA · EADs->HMDB51 split 1 | accuracy: 50.5 -> 52.3 | Improved |
| add stages 3-4 (cycle back to image model) instead of one video self-training round (ablation) | CycDA · EADs->HMDB51 split 1 | accuracy: 55.4 -> 60.8 | Improved |
| use two rounds of video self-training instead of the full cycle (ablation) | CycDA · EADs->HMDB51 split 1 | accuracy: 56.4 -> 60.8 | Improved |
| add source data to stage-3 supervision (ablation) | CycDA · EADs->HMDB51 split 1 | accuracy: 55.6 -> 56.4 | Improved |
| add adversarial domain discrimination in stage 3 (ablation) | CycDA · EADs->HMDB51 split 1 | accuracy: 56.4 -> 58.0 | Improved |
| use cross-domain contrastive learning in stage 3 instead of adversarial (ablation) | CycDA · EADs->HMDB51 split 1 | accuracy: 58.0 -> 60.8 | Improved |
| add more CycDA iterations | CycDA · EADs->HMDB51 | accuracy: 60.5 -> 62.0 | Mixed |
| add 5% source videos to web images | CycDA · UCF-HMDB U->H | accuracy: 77.8 -> 82.2 | Improved |
| increase web images per class | CycDA · UCF-HMDB U->H | accuracy: 77.8 -> 82.2 | Improved |
| add 5 frames per source video instead of 1 | CycDA · UCF-HMDB U->H | accuracy: 83.3 -> 84.4 | Improved |
Closest papers by meaning
- Source-Free Video Domain Adaptation With Spatial-Temporal-Historical Consistency Learning (CVPR 2023 · similarity 0.89)
- Unsupervised Video Domain Adaptation for Action Recognition: A Disentanglement Perspective (NeurIPS 2023 · similarity 0.89)
- Source-Free Video Domain Adaptation by Learning Temporal Consistency for Action Recognition (ECCV 2022 · similarity 0.89)
- Unsupervised Video Domain Adaptation with Masked Pre-Training and Collaborative Self-Training (CVPR 2024 · similarity 0.88)
- Multi-Modal Domain Adaptation for Fine-Grained Action Recognition (CVPR 2020 · similarity 0.88)
- Modality-Collaborative Test-Time Adaptation for Action Recognition (CVPR 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.
