Research map / Video / Action recognition
Flip-Invariant Motion Representation
Takumi Kobayashi
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 |
|---|---|---|---|
| estimate patch orientation from flow instead of image gradients (Flip by Flow) (ablation) | HOF descriptor · HMDB51 | accuracy (%): 52.24 -> 52.81 | Improved |
| estimate patch flip from flow instead of image gradients (ablation) | HOG descriptor · Hollywood2 | accuracy (%): 48.15 -> 50.37 | Improved |
| add flip augmentation of training and test samples (ablation) | HOF descriptor · HMDB51 | accuracy (%): 50.00 -> 49.59 | No clear effect |
| add flip invariance by Flip by Flow (ablation) | HOF descriptor · UCF101 | accuracy (%): 77.44 -> 79.27 | Improved |
| use learned eigen invariant form (5) instead of averaging (ablation) | HOF descriptor · HMDB51 | accuracy (%): 52.24 -> 54.64 | Improved |
| use eigen invariant form instead of MI-SIFT (ablation) | HOF descriptor · UCF101 | accuracy (%): 77.24 -> 80.09 | Improved |
| add orthonormality constraint to the learned invariant form (vs generalized eigenvalue problem) (ablation) | temporal ConvNet layer4 · HMDB51 | accuracy (%): 56.45 -> 57.30 | Improved |
| use MI-SIFT on ConvNet descriptors (ablation) | temporal ConvNet layer3 · Hollywood2 | accuracy (%): 62.72 -> 60.50 | Hurt |
| use max operator instead of average for invariance (ablation) | temporal ConvNet layer4 · UCF101 | accuracy (%): 84.47 -> 85.14 | Improved |
| add learned invariant form (ablation) | spatial ConvNet layer4 · HMDB51 | accuracy (%): 46.67 -> 46.41 | No clear effect |
| train invariant form on UCF101 and apply to other datasets (ablation) | temporal ConvNet layer4 · HMDB51 | accuracy (%): 57.30 -> 56.80 | No clear effect |
| add Flip by Flow to descriptor combination | iDT+LMS+ConvNet · Hollywood2 | accuracy (%): 70.53 -> 72.49 | Improved |
Closest papers by meaning
- Recognize Actions by Disentangling Components of Dynamics (CVPR 2018 · similarity 0.87)
- Learning Video Representations by Transforming Time (ECCV 2020 · similarity 0.85)
- Unsupervised Learning of View-invariant Action Representations (NeurIPS 2018 · similarity 0.84)
- DVANet: Disentangling View and Action Features for Multi-View Action Recognition (AAAI 2024 · similarity 0.84)
- DisMo: Disentangled Motion Representations for Open-World Motion Transfer (NeurIPS 2025 · similarity 0.83)
- Disentangled Concepts Speak Louder Than Words: Explainable Video Action Recognition (NeurIPS 2025 · similarity 0.82)
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.
