Research map / Video / Video segmentation and tracking
Greedy Hierarchical Variational Autoencoders for Large-Scale Video Prediction
Check 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 greedy hierarchical VAE modules instead of SVG' | video prediction · RoboNet | FVD: 123.2$\pm$2.6 -> 95.2$\pm$2.6 | Helped |
| use greedy hierarchical VAE modules instead of SVG' | video prediction · KITTI | FVD: 1217.3~[cite] -> 552.9$\pm$21.2 | Helped |
| use greedy hierarchical VAE modules instead of SVG' | video prediction · Human3.6M | FVD: 429.9~[cite] -> 355.2$\pm$2.9 | Helped |
| use greedy hierarchical VAE modules instead of Hier-VRNN | video prediction · Cityscapes | FVD: 567.5~[cite] -> 418.0$\pm$5.0 | Helped |
| use GHVAEs instead of SVG' | robot planning · Pick&Wipe | success rate: 50.0\% -> 90.0\% | Helped |
| use GHVAEs instead of SVG' | robot planning · Pick&Sweep | success rate: 50.0\% -> 85.0\% | Helped |
| add modules, increasing the stack from 2 to 6 (ablation) | GHVAE · RoboNet | FVD: 292.4$\pm$11.1 -> 95.2$\pm$2.6 | Helped |
| train all modules end-to-end from scratch instead of greedily (ablation) | GHVAE · RoboNet | FVD: 95.2$\pm$2.6 -> 509.9$\pm$6.2 | Hurt |
| fine-tune a greedily trained model end-to-end (ablation) | GHVAE · RoboNet | FVD: 95.2$\pm$2.6 -> 91.1$\pm$3.1 | Helped |
| replace a learned prior with a uniform prior (ablation) | GHVAE · RoboNet | FVD: 95.2$\pm$2.6 -> 281.4$\pm$1.6 | Hurt |
| replace the spatially compressing, channel-expanding encoder with an architecture that keeps or expands spatial dimensions while reducing channels (ablation) | GHVAE · video prediction | prediction performance | Hurt |
| use all module latents at test time instead of only the deepest latent (ablation) | GHVAE · video prediction | prediction performance | Hurt |
Closest papers by meaning
- Clockwork Variational Autoencoders (NeurIPS 2021 · similarity 0.88)
- Deep Hierarchical Video Compression (AAAI 2024 · similarity 0.88)
- Improved Video VAE for Latent Video Diffusion Model (CVPR 2025 · similarity 0.86)
- EVEREST: Efficient Masked Video Autoencoder by Removing Redundant Spatiotemporal Tokens (ICML 2024 · similarity 0.85)
- $t^3$-Variational Autoencoder: Learning Heavy-tailed Data with Student's t and Power Divergence (ICLR 2024 · similarity 0.85)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.85)
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.