Research map / Video / Video segmentation and tracking
MMVP: Motion-Matrix-Based Video Prediction
Yiqi Zhong, Luming Liang, Ilya Zharkov, Ulrich Neumann
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 the filter block before motion matrix construction (ablation) | MMVP · UCF Sports (own split, hard subset) | LPIPS: 0.3819 -> 0.3510 | Improved |
| include more feature scales in future composition (add 1/2 scale) (ablation) | MMVP · UCF Sports (own split, full set) | LPIPS: 0.1874 -> 0.1115 | Improved |
| add the full-resolution (scale 1) features to composition (ablation) | MMVP · UCF Sports (own split, full set) | LPIPS: 0.1115 -> 0.1064 | Improved |
| add the input image as a composition source (ablation) | MMVP · UCF Sports (own split, full set) | LPIPS: 0.1064 -> 0.1051 | Improved |
| drop the largest-scale features when the image is already used (ablation) | MMVP · UCF Sports (own split, full set) | LPIPS: 0.1051 -> 0.1062 | No clear effect |
| increase motion feature length from 4 to 8 (with image length 16) (ablation) | MMVP · UCF Sports (own split, full set) | LPIPS: 0.1184 -> 0.1175 | Improved |
| predict motion matrices from last observed frame to each future frame instead of consecutive frames | MMVP · KTH 10->20 vs 10->40 | PSNR: 27.54 vs 26.35 (small gap between horizons) | Improved |
| use two-stream MMVP instead of single-stream SimVP on harder samples | SimVP · UCF Sports (own split, hard subset) | LPIPS: 0.5600 -> 0.3510 | Improved |
| apply MMVP on Moving MNIST versus best baselines | CrevNet · Moving MNIST | MSE: 22.3 -> 22.2 | No clear effect |
| decouple motion from appearance so prediction modules handle only motion | STIP/SimVP style predictors · UCF Sports | parameters: 18054.2K -> 2789.0K total | Improved |
Later papers that compared with this method
2 later accepted papers report MMVP as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- DFDNet: Disentangling and Filtering Dynamics for Enhanced Video Prediction (AAAI 2025 · better on 2 of 2 shared results)
- Motion Graph Unleashed: A Novel Approach to Video Prediction (NeurIPS 2024 · better on 9 of 9 shared results)
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.
- SimVP — SimVP: Simpler Yet Better Video Prediction (CVPR 2022) · better on 3 of 3 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
- OCK: Unsupervised Dynamic Video Prediction with Object-Centric Kinematics (ICCV 2025 · similarity 0.86)
- Motion Graph Unleashed: A Novel Approach to Video Prediction (NeurIPS 2024 · similarity 0.86)
- Video Frame Prediction from a Single Image and Events (AAAI 2024 · similarity 0.85)
- VideoJAM: Joint Appearance-Motion Representations for Enhanced Motion Generation in Video Models (ICML 2025 · similarity 0.84)
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.84)
- HumMUSS: Human Motion Understanding using State Space Models (CVPR 2024 · similarity 0.83)
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.
