Research map / Probabilistic and scientific modeling / Variational and generative sampling
Improved Conditional VRNNs for Video Prediction
Lluis Castrejon, Nicolas Ballas, Aaron Courville
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 |
|---|---|---|---|
| add ConvLSTM layers to the decoder (1 to 3) (ablation) | VRNN likelihood · BAIR Push | test ELBO: 3826 -> 2355 | Improved |
| add ConvLSTM layers to the decoder (3 to 6) (ablation) | VRNN likelihood · BAIR Push | test ELBO: 2355 -> 1731.31 | Improved |
| increase ConvLSTM hidden size of the 6-layer decoder (ablation) | VRNN likelihood · BAIR Push | test ELBO: 1731.31 -> 1589.72 | Improved |
| add a second latent level at 8x8 (ablation) | VRNN with 1 latent level · BAIR Push | test ELBO: 1536.93 -> 1313.02 | Improved |
| add a third latent level at 32x32 (1-8-32) (ablation) | VRNN with 2 latent levels · BAIR Push | test ELBO: 1313.02 -> 1162.24 | Improved |
| add a fourth latent level at 16x16 (1-8-16-32) (ablation) | 1-8-32 VRNN · BAIR Push | test ELBO: 1162.24 -> 1256.22 | Hurt |
| add beta warmup and dense connectivity together (ablation) | 1-8-32 VRNN · BAIR Push | test ELBO: 1440.58 -> 1162.24 | Improved |
| add beta warmup alone (ablation) | 1-8-32 VRNN · BAIR Push | test ELBO: 1440.58 -> 1440.62 | No clear effect |
| add dense connectivity alone (ablation) | 1-8-32 VRNN · BAIR Push | test ELBO: 1440.58 -> 1547.05 | Hurt |
| add latent hierarchy to single-level VRNN | Ours · Stochastic Moving MNIST | FVD: 63.81 -> 57.17 | Improved |
| add latent hierarchy to single-level VRNN | Ours · Stochastic Moving MNIST | SSIM: 0.763 -> 0.760 | No clear effect |
| replace SVG-LP with higher-capacity VRNN | SVG-LP · Stochastic Moving MNIST | FVD: 90.81 -> 63.81 | Improved |
Closest papers by meaning
- Deep Hierarchical Video Compression (AAAI 2024 · similarity 0.88)
- CV-VAE: A Compatible Video VAE for Latent Generative Video Models (NeurIPS 2024 · similarity 0.87)
- VideoFlow: A Conditional Flow-Based Model for Stochastic Video Generation (ICLR 2020 · similarity 0.87)
- Probabilistic Video Generation using Holistic Attribute Control (ECCV 2018 · similarity 0.87)
- VEDIT: Latent Prediction Architecture For Procedural Video Representation Learning (ICLR 2025 · similarity 0.87)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.86)
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.
