Research map / Video / Video segmentation and tracking
MOSO: Decomposing MOtion, Scene and Object for Video Prediction
Mingzhen Sun, Weining Wang, Xinxin Zhu, Jing Liu
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 |
|---|---|---|---|
| decompose video into motion, scene, and object tokens and predict at token level | MOSO · BAIR video prediction | FVD: 86.9 -> 83.6 | Improved |
| decompose video into motion, scene, and object tokens and predict at token level | MOSO · RoboNet 64² video prediction | FVD: 62.5 -> 53.3 | Improved |
| decompose video into motion, scene, and object tokens and predict at token level | MOSO · KITTI 256² video prediction | FVD: 446.1 -> 516.6 | Hurt |
| decompose video into motion, scene, and object tokens and predict at token level | MOSO · KTH 64² video prediction | PSNR: 27.1 -> 29.8 | Improved |
| use motion, scene, and object decomposition instead of no decomposition (ablation) | MOSO-VQVAE · KTH video reconstruction | PSNR: 24.8 -> 36.0 | Improved |
| separate objects from scenes rather than treating them as combined content (ablation) | MOSO-VQVAE · KTH video reconstruction | PSNR: 32.6 -> 36.0 | Improved |
| add the preprocessing algorithm (ablation) | MOSO-VQVAE · UCF101 video reconstruction | FVD: 310.1 -> 267.9 | Improved |
| remove the preprocessing algorithm (ablation) | MOSO · KTH 64² video prediction | PSNR: 29.8 -> 29.2 | Hurt |
| remove the merge module (ablation) | MOSO · KTH 64² video prediction | LPIPS: 0.083 -> 0.125 | Hurt |
| use a single Transformer instead of the MOSO Transformer arrangement (ablation) | MOSO · KTH 64² video prediction | LPIPS: 0.083 -> 0.116 | Hurt |
| add a video discriminator with loss weight 0.1 (ablation) | MOSO-VQVAE · UCF101 video reconstruction | FID: 24.8 -> 17.9 | Improved |
| add an image discriminator alongside the video discriminator (ablation) | MOSO-VQVAE · UCF101 video reconstruction | FID: 17.9 -> 19.5 | Hurt |
Later papers that compared with this method
2 later accepted papers report MOSO as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- LaMD: Latent Motion Diffusion for Image-Conditional Video Generation (IJCV 2025 · better on 1 of 1 shared result)
- GLOBER: Coherent Non-autoregressive Video Generation via GLOBal Guided Video DecodER (NeurIPS 2023 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- MaskViT — MaskViT: Masked Visual Pre-Training for Video Prediction (ICLR 2023) · better on 2 of 3 shared results
- StyleGAN-V — StyleGAN-V: A Continuous Video Generator With the Price, Image Quality and Perks of StyleGAN2 (CVPR 2022) · better on 1 of 1 shared result
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
- Moving Off-the-Grid: Scene-Grounded Video Representations (NeurIPS 2024 · similarity 0.86)
- Video Diffusion Models are Training-free Motion Interpreter and Controller (NeurIPS 2024 · similarity 0.83)
- Through-The-Mask: Mask-based Motion Trajectories for Image-to-Video Generation (CVPR 2025 · similarity 0.83)
- Enhancing Motion in Text-to-Video Generation with Decomposed Encoding and Conditioning (NeurIPS 2024 · similarity 0.83)
- Decouple Content and Motion for Conditional Image-to-Video Generation (AAAI 2024 · similarity 0.83)
- DreamScene4D: Dynamic Multi-Object Scene Generation from Monocular Videos (NeurIPS 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.
