Research map / Video / Video segmentation and tracking
A Dynamic Multi-Scale Voxel Flow Network for Video Prediction
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 DMVFN instead of OPT | Video prediction · Cityscapes-Test | MS-SSIM (×10⁻²), t+1: 94.54 -> 95.73 | Helped |
| Use DMVFN instead of OPT | Video prediction · KITTI-Test | MS-SSIM (×10⁻²), t+5: 61.09 -> 70.52 | Helped |
| Use DMVFN instead of OPT | Video prediction · Vimeo-Test | LPIPS (×10⁻²): 3.59 -> 3.69 | Hurt |
| Add dynamic routing (ablation) | DMVFN · Cityscapes-Test | GFLOPs: 24.51 -> 12.71 | Helped |
| Add dynamic routing (ablation) | DMVFN · Cityscapes-Test | MS-SSIM (×10⁻²), t+5: 81.48 -> 83.45 | Helped |
| Use STEBS rather than Gumbel Softmax routing (ablation) | DMVFN · Cityscapes | MS-SSIM (×10⁻²), t+5: 79.54 -> 83.45 | Helped |
| Use multiscale [4,2,1] rather than [1] (ablation) | DMVFN · Cityscapes | MS-SSIM (×10⁻²), t+5: 80.93 -> 83.45 | Helped |
| Add the spatial path (ablation) | DMVFN · KITTI | MS-SSIM (×10⁻²), t+5: 69.86 -> 70.52 | Helped |
| Increase inference β from 0.3 to 0.8 (ablation) | DMVFN · KITTI | MS-SSIM (×10⁻²): 78.78 -> 88.89 | Helped |
| Increase inference β from 0.3 to 0.8 (ablation) | DMVFN · KITTI | GFLOPs: 2.62 -> 6.40 | Hurt |
| Use intermediate-output losses rather than final-output-only supervision (ablation) | DMVFN · KITTI | MS-SSIM (×10⁻²), t+5: 70.18 -> 70.52 | Helped |
| Use DMVFN rather than STRPM | Video prediction · Human3.6M, t+1 | PSNR: 33.32 -> 35.07 | Helped |
Closest papers by meaning
- MVFNet: Multi-View Fusion Network for Efficient Video Recognition (AAAI 2021 · similarity 0.86)
- DFDNet: Disentangling and Filtering Dynamics for Enhanced Video Prediction (AAAI 2025 · similarity 0.85)
- Dynamic Video Segmentation Network (CVPR 2018 · similarity 0.85)
- Dynamic Filter Networks (NeurIPS 2016 · similarity 0.84)
- High Fidelity Video Prediction with Large Stochastic Recurrent Neural Networks (NeurIPS 2019 · similarity 0.84)
- 3DV: 3D Dynamic Voxel for Action Recognition in Depth Video (CVPR 2020 · 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.