Research map / Label-efficient and robust learning / Domain adaptation and generalization
Augmenting and Aligning Snippets for Few-Shot Video Domain Adaptation
Yuecong Xu, Jianfei Yang, Yunjiao Zhou, Zhenghua Chen, Min Wu, Xiaoli Li
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 Stochastic Sampling Augmentation of snippets (ablation) | SSA2lign · Daily-DA/Sports-DA (5 tasks) | average top-1 accuracy: 58.679 -> 63.425 | Improved |
| apply SSA with prediction loss only, no adaptation (ablation) | TSF w/ T · Daily-DA/Sports-DA (5 tasks) | average top-1 accuracy: 53.859 -> 55.734 | Improved |
| use random frame-level spatial augmentation instead of snippet sampling (ablation) | SSA2lign · Daily-DA/Sports-DA (5 tasks) | average top-1 accuracy: 63.425 -> 58.424 | Hurt |
| add statistical snippet alignment (L_sn-stat) (ablation) | SSA2lign · Daily-DA/Sports-DA (5 tasks) | average top-1 accuracy: 62.798 -> 63.425 | Improved |
| add snippet attention (ablation) | SSA2lign · Daily-DA/Sports-DA (5 tasks) | average top-1 accuracy: 62.455 -> 62.798 | Improved |
| compute cross-snippet consistency over all snippet pairs (ablation) | SSA2lign · Daily-DA/Sports-DA (5 tasks) | average top-1 accuracy: 63.425 -> 63.769 | No clear effect |
| align target snippets to source class prototypes instead of target prototypes (ablation) | SSA2lign · Daily-DA/Sports-DA (5 tasks) | average top-1 accuracy: 63.425 -> 62.630 | Hurt |
| swap MDD for CORAL or MMD as statistical metric (ablation) | SSA2lign · Daily-DA/Sports-DA (5 tasks) | average top-1 accuracy: 63.425 -> 63.160 (CORAL), 63.072 (MMD) | No clear effect |
| use more than one snippet per target video (r>=3) (ablation) | SSA2lign · U->S | accuracy | Improved |
| train prior FSDA and UDA/VUDA methods on few-shot target data versus prediction loss only | FADA, DANN, TRX, etc. · Daily-DA 10-shot | top-1 accuracy | Hurt |
| add few target videos with prediction loss only at k=3 | TimeSFormer · Sports-DA 3-shot | top-1 accuracy: 84.953 -> 84.652 | Mixed |
| vary lambda_sem:lambda_stat ratio and lambda_P (ablation) | SSA2lign · U->S | accuracy: within 0.683% | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- d-SNE — d-SNE: Domain Adaptation Using Stochastic Neighborhood Embedding (CVPR 2019) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Source-Free Video Domain Adaptation With Spatial-Temporal-Historical Consistency Learning (CVPR 2023 · similarity 0.90)
- The Unreasonable Effectiveness of Large Language-Vision Models for Source-Free Video Domain Adaptation (ICCV 2023 · similarity 0.89)
- Learnable Motion-Focused Tokenization for Effective and Efficient Video Unsupervised Domain Adaptation (CVPR 2026 · similarity 0.88)
- Align Before Adapt: Leveraging Entity-to-Region Alignments for Generalizable Video Action Recognition (CVPR 2024 · similarity 0.87)
- Unified Language-driven Zero-shot Domain Adaptation (CVPR 2024 · similarity 0.87)
- Modality-Collaborative Test-Time Adaptation for Action Recognition (CVPR 2024 · 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.
