Research map / Image generation / Autoregressive and latent generation
Meta-Sim: Learning to Generate Synthetic Datasets
Amlan Kar, Aayush Prakash, Ming-Yu Liu, Eric Cameracci, Justin Yuan, Matt Rusiniak, David Acuna, Antonio Torralba, Sanja Fidler
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 Meta-Sim distribution transformer | probabilistic grammar · MNIST (rotation gap) | accuracy: 14.8 -> 99.5 | Improved |
| add Meta-Sim distribution transformer | probabilistic grammar · MNIST (rotation + translation gap) | accuracy: 13.1 -> 99.3 | Improved |
| add Meta-Sim distribution transformer | probabilistic grammar · Aerial2D segmentation | mean IoU: 80.3 -> 95.2 | Improved |
| add Meta-Sim distribution transformer | probabilistic grammar · Aerial2D segmentation | car IoU: 30.0 -> 86.7 | Improved |
| add Meta-Sim distribution transformer | probabilistic grammar · Aerial2D segmentation | house IoU: 98.3 -> 95.0 | Hurt |
| add Meta-Sim car attribute optimization | probabilistic grammar · KITTI val car detection (Easy) | AP@0.5 IoU: 63.7 -> 66.4 | Improved |
| add camera parameters to optimization (ablation) | MetaSim (Cars) · KITTI val car detection (Moderate) | AP@0.5 IoU: 66.5 -> 66.3 | No clear effect |
| add context elements to optimization (ablation) | MetaSim + Camera · KITTI val car detection (Hard) | AP@0.5 IoU: 65.9 -> 66.0 | No clear effect |
| add meta task loss after MMD (ablation) | MetaSim + Context · KITTI val car detection (Easy) | AP@0.5 IoU: 65.9 -> 66.7 | Improved |
| add MUNIT image translation | Meta-Sim · KITTI val car detection (Easy) | AP@0.5 IoU: 71.1 -> 77.5 | Improved |
| add MUNIT image translation, compare Meta-Sim to grammar | Meta-Sim · KITTI val car detection (Moderate) | AP@0.5 IoU: 75.5 -> 75.1 | Hurt |
| finetune on V after Meta-Sim pretraining | Mask-RCNN · KITTI val car detection (Moderate) | AP@0.5 IoU: 72.7 -> 73.9 | Improved |
Closest papers by meaning
- The Unmet Promise of Synthetic Training Images: Using Retrieved Real Images Performs Better (NeurIPS 2024 · similarity 0.87)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.86)
- Generative Photography: Scene-Consistent Camera Control for Realistic Text-to-Image Synthesis (CVPR 2025 · similarity 0.85)
- DreamText: High Fidelity Scene Text Synthesis (CVPR 2025 · similarity 0.85)
- SimGen: Simulator-conditioned Driving Scene Generation (NeurIPS 2024 · similarity 0.85)
- Context-Aware Meta-Learning (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.
