Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised Real-to-Sim Scene Generation
Aayush Prakash, Shoubhik Debnath, Jean-Francois Lafleche, Eric Cameracci, Gavriel State, Stan Birchfield, Marc T. Law
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 label alignment sigma^c,label (ablation) | Sim2SG with appearance and prediction alignment · KITTI hard | mAP@0.5 IoU: 0.246 -> 0.316 | Improved |
| add label alignment sigma^c,label (ablation) | Sim2SG with appearance and prediction alignment · KITTI hard | relationship triplet Recall@50: 0.076 -> 0.139 | Improved |
| add appearance alignment sigma^a to label alignment (ablation) | Sim2SG · KITTI hard | mAP@0.5 IoU: 0.285 -> 0.311 | Improved |
| add appearance alignment sigma^a to label alignment (ablation) | Sim2SG · KITTI hard | relationship triplet Recall@50: 0.104 -> 0.127 | Improved |
| add prediction alignment sigma^c,pred (ablation) | Sim2SG with label and appearance alignment · KITTI hard | relationship triplet Recall@50: 0.127 -> 0.139 | Improved |
| add appearance and prediction alignment to label alignment (ablation) | Sim2SG · KITTI hard, Pedestrian AP | AP@0.5 IoU: 0.262 -> 0.241 | Hurt |
| use Sim2SG instead of SDR synthetic data | detector trained on synthetic data · KITTI hard | mAP@0.5 IoU: 0.234 -> 0.316 | Improved |
| use Sim2SG instead of Meta-Sim | scene generation · KITTI hard | relationship triplet Recall@50: 0.075 -> 0.139 | Improved |
| use GPA feature alignment instead of SDR only | domain adaptation baseline · KITTI hard | mAP@0.5 IoU: 0.234 -> 0.087 | Hurt |
| add appearance alignment sigma^a to label alignment (ablation) | Sim2SG · CLEVR target domain | relationship triplet Recall@20: 0.493 -> 0.888 | Improved |
| add appearance alignment sigma^a in a content-gap-only setting (ablation) | Sim2SG vs SDR · CLEVR (same appearance, different content) | relationship triplet Recall@20: 0.760 -> 0.722 | Hurt |
| add label alignment sigma^c,label in a content-gap-only setting (ablation) | Sim2SG vs SDR · CLEVR (same appearance, different content) | relationship triplet Recall@20: 0.76 -> 0.996 | Improved |
Closest papers by meaning
- Meta-Sim: Learning to Generate Synthetic Datasets (ICCV 2019 · similarity 0.90)
- Meta-Sim2: Unsupervised Learning of Scene Structure for Synthetic Data Generation (ECCV 2020 · similarity 0.89)
- Adversarially Tuned Scene Generation (CVPR 2017 · similarity 0.88)
- A New Benchmark: On the Utility of Synthetic Data With Blender for Bare Supervised Learning and Downstream Domain Adaptation (CVPR 2023 · similarity 0.87)
- Generate Any Scene: Scene Graph Driven Data Synthesis for Visual Generation Training (ICLR 2026 · similarity 0.87)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · 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.
