Reproduce “Motion Graph Unleashed: A Novel Approach to Video Prediction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yiqi Zhong, Luming Liang, Bohan Tang, Ilya Zharkov, Ulrich Neumann
This paper proposes a graph-based motion representation for video prediction, constructing a motion graph where video patches are nodes connected by spatial and temporal edges. The approach enables explicit modeling of complex motion dependencies and achieves competitive performance on datasets like UCF Sports, KITTI, and Cityscapes, while significantly reducing model size and GPU memory usage.
Key insights
- A graph structure can effectively model complex spatial-temporal relationships between video patches, improving motion prediction.
- Representing motion at a patch-graph level provides a balance between detailed motion capture and computational efficiency.
- Iterative message passing over spatial and temporal edges within the graph enables robust information propagation for motion learning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.