Research map / Video / Action recognition
Aligning Videos in Space and Time
Senthil Purushwalkam, Tian Ye, Saurabh Gupta, Abhinav Gupta
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 |
|---|---|---|---|
| train with ground-truth keypoint tracks instead of tracker tracks (ablation) | cross-video cycle-consistency features · Penn Action track correspondence | accuracy: 0.551 -> 0.650 | Improved |
| add temporal alignment search during training (ablation) | cross-video cycle-consistency features (GT tracks) · Penn Action track correspondence | accuracy: 0.584 -> 0.650 | Improved |
| add reference video retrieval before alignment (ablation) | cross-video cycle-consistency features · Penn Action track correspondence | accuracy: drops by 15% | Improved |
| replace ImageNet features with cross-video cycle features | ResNet-18 · Penn Action temporal alignment | alignment error: 0.509 -> 0.448 | Improved |
| replace ImageNet features with cross-video cycle features | ResNet-18 · Penn Action spatial alignment | accuracy: 0.153 -> 0.284 | Improved |
| replace ImageNet features with cross-video cycle features | ResNet-18 · Penn Action track correspondence | accuracy: 0.252 -> 0.551 | Improved |
| replace Penn Action action classification features with cycle features | ResNet-18 · Penn Action track correspondence | accuracy: 0.110 -> 0.551 | Improved |
| replace intra-video tracking features with cross-video cycle features | ResNet-18 · Penn Action spatial alignment | accuracy: 0.060 -> 0.284 | Improved |
| use soft max over cycle scores instead of hard max (ablation) | cycle-consistency loss · training | loss optimization | Improved |
Closest papers by meaning
- Learning by Aligning Videos in Time (CVPR 2021 · similarity 0.90)
- Representation Learning via Global Temporal Alignment and Cycle-Consistency (CVPR 2021 · similarity 0.89)
- Temporal Alignment Networks for Long-Term Video (CVPR 2022 · similarity 0.89)
- Align Before Adapt: Leveraging Entity-to-Region Alignments for Generalizable Video Action Recognition (CVPR 2024 · similarity 0.88)
- Temporal Alignment-Free Video Matching for Few-shot Action Recognition (CVPR 2025 · similarity 0.87)
- VideoComp: Advancing Fine-Grained Compositional and Temporal Alignment in Video-Text Models (CVPR 2025 · 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.
