Research map / Video / Video segmentation and tracking
Understanding image motion with group representations
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use image pairs as CNN input rather than single images (ablation) | Motion representation model · MNIST and KITTI held-out sequences | Embedding error: MNIST equivalent: 0.74 -> 8.1\mathrm{e}{-4}; MNIST inequivalent: 0.79 -> 1.7\mathrm{e}{-2}; KITTI equivalent: 3.5\mathrm | Helped |
| Train a global motion embedding with group-property constraints | Motion representation model · KITTI visual odometry test | Translation X MSE (cm): Mean 5.92 -> Ours 5.05 | Helped |
| Train an embedding to represent typical sequence dynamics | Motion representation model · KITTI interpolation, 1 skipped frame | Distance to true versus alternative middle frame: True 3.91e-03; inside 7.67e-02; outside 2.94e-01 | Helped |
| Replace the learned embedding distance with pixelwise Euclidean distance | KITTI interpolation evaluation · KITTI interpolation, 1 skipped frame | Distance discrimination for the true middle frame: True 7.92e-04; inside 7.97e-04; outside 1.09e-03 | Hurt |
| Use Flow+PCA (256 PCs) rather than the learned global embedding | KITTI visual odometry regression · KITTI visual odometry test | Translation Z MSE (cm): Ours 1539.04 ± 2.3e+01 (19.20%) -> Flow+PCA 239.46 ± 5.6e+00 (87.43%) | Mixed |
Closest papers by meaning
- SEPT: Towards Efficient Scene Representation Learning for Motion Prediction (ICLR 2024 · similarity 0.85)
- Large Displacement Motion Transfer with Unsupervised Anytime Interpolation (ICML 2025 · similarity 0.85)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.85)
- ProMotion: Prototypes As Motion Learners (CVPR 2024 · similarity 0.84)
- MoST: Multi-Modality Scene Tokenization for Motion Prediction (CVPR 2024 · similarity 0.84)
- DisMo: Disentangled Motion Representations for Open-World Motion Transfer (NeurIPS 2025 · similarity 0.84)
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.