Research map / 3D vision / 3D generation
Attend, Infer, Repeat: Fast Scene Understanding with Generative Models
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 DRAW with 40 steps instead of AIR with up to 3 steps | AIR · Multi-MNIST up to 2 digits | free energy: -620 -> -637 | Helped |
| use DAIR instead of DRAW on 3-digit generalization | DRAW · Multi-MNIST only 3 digits | free energy: -406 -> -424 | Helped |
| use DAIR (difference-based recurrent architecture) instead of AIR for extrapolation (ablation) | AIR · Multi-MNIST only 3 digits | free energy: -316 -> -424 | Helped |
| AIR trained on up to 2 digits tested on 3 digits count | AIR · Multi-MNIST extrapolation | count accuracy | Hurt |
| use DAIR instead of AIR for count generalization | AIR · Multi-MNIST extrapolation/interpolation | count accuracy | Helped |
| use AIR inferences as features vs raw-pixel CNN for digit sum and order | CNN · Multi-MNIST downstream tasks | accuracy | Helped |
| use AIR embedding instead of CAE or DRAW embedding | CAE/DRAW · Multi-MNIST downstream tasks | accuracy | Helped |
| unsupervised AIR instead of supervised prediction on 3D single objects | supervised network · 3D objects cube pose | reconstruction quality | Helped |
| amortized AIR inference instead of direct gradient descent per image | direct optimization · 3D objects | stability | Helped |
| AIR instead of supervised network on 3D tabletop scenes with repetitions | supervised network · 3D tabletop scenes | reconstruction and count accuracy | Helped |
| AIR vs supervised on tabletop scenes without repetitions | supervised network · 3D tabletop scenes | reconstruction and count accuracy | No clear effect |
Closest papers by meaning
- Faster Attend-Infer-Repeat with Tractable Probabilistic Models (ICML 2019 · similarity 0.91)
- Unsupervised Learning of 3D Structure from Images (NeurIPS 2016 · similarity 0.89)
- GENESIS: Generative Scene Inference and Sampling with Object-Centric Latent Representations (ICLR 2020 · similarity 0.89)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.88)
- Data-efficient Large Vision Models through Sequential Autoregression (ICML 2024 · similarity 0.86)
- DRScaffold: Boosting Dense-Scene Reasoning in Lightweight Vision Language Models (NeurIPS 2026 · 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.