Research map / Video / Video segmentation and tracking
VideoFlow: A Conditional Flow-Based Model for Stochastic Video Generation
Manoj Kumar, Mohammad Babaeizadeh, Dumitru Erhan, Chelsea Finn, Sergey Levine, Laurent Dinh, Durk Kingma
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 |
|---|---|---|---|
| Use VideoFlow instead of SAVP-VAE | stochastic video generation · Stochastic Movement Dataset | fooling rate: 16.4 % -> 31.8 % | Improved |
| Use VideoFlow instead of SV2P | stochastic video generation · Stochastic Movement Dataset | fooling rate: 17.5 % -> 31.8 % | Improved |
| Use VideoFlow instead of SAVP-VAE | conditional video prediction · BAIR action-free | bits-per-pixel: ≤ 6.73 -> 1.87 | Improved |
| Use VideoFlow instead of SV2P | conditional video prediction · BAIR action-free | bits-per-pixel: ≤ 6.78 -> 1.87 | Improved |
| Lower sampling temperature from T=1.0 to T=0.8 | VideoFlow · BAIR action-free, training 13 frames; evaluation 13 frames | FVD: 149$\pm$6 -> 95$\pm$4 | Improved |
| Lower sampling temperature from T=1.0 to T=0.8 | VideoFlow · BAIR action-free, training 13 frames; evaluation 16 frames | FVD: 221$\pm$8 -> 127$\pm$3 | Improved |
| Lower sampling temperature from T=1.0 to T=0.8 | VideoFlow · BAIR action-free, training 13 frames; 2 conditioning frames; evaluation 16 frames | FVD: 251$\pm$7 -> 131$\pm$5 | Improved |
| Add temporal skip connections (ablation) | VideoFlow latent dynamics network · BAIR action-free | bits-per-pixel | Improved |
| Use dilated convolutions and GATU instead of the baseline components (ablation) | VideoFlow latent dynamics network · BAIR action-free | bits-per-pixel | Improved |
| Add dilated convolutions and temporal skip connections (ablation) | VideoFlow latent dynamics network · BAIR action-free | bits-per-pixel | Improved |
| Use dilated convolutions alone, without temporal skip connections or GATU (ablation) | VideoFlow latent dynamics network · BAIR action-free | training stability | Hurt |
| Lower temperature from 1.0 toward 0.0 (ablation) | SAVP-VAE and SV2P latent Gaussian priors · BAIR action-free | performance | Hurt |
Later papers that compared with this method
2 later accepted papers report VideoFlow as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- H-MoRe: Learning Human-centric Motion Representation for Action Analysis (CVPR 2025 · better on 3 of 3 shared results)
- Scaling Autoregressive Video Models (ICLR 2020 · 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
- STARFlow-V: End-to-End Video Generative Modeling with Autoregressive Normalizing Flows (CVPR 2026 · similarity 0.91)
- Flow-Grounded Spatial-Temporal Video Prediction from Still Images (ECCV 2018 · similarity 0.91)
- Stochastic Video Generation with a Learned Prior (ICML 2018 · similarity 0.89)
- Taming generative video models for zero-shot optical flow extraction (NeurIPS 2025 · similarity 0.88)
- StreamFlow: Streamlined Multi-Frame Optical Flow Estimation for Video Sequences (NeurIPS 2024 · similarity 0.87)
- MESSENGER: Memory-Enhanced Sequential Scene Flow Estimation via Autoregressive Next-Frame Forecasting (NeurIPS 2026 · similarity 0.87)
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.
