Research map / Vision-language and multimodal / Human and embodied modeling
Imagine the Unseen World: A Benchmark for Systematic Generalization in Visual World Models
Yeongbin Kim, Gautam Singh, Junyeong Park, Caglar Gulcehre, Sungjin Ahn
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 |
|---|---|---|---|
| use ViT encoder instead of CNN encoder | Image-to-Image model · SVIB-dSprites Easy | LPIPS: 0.0643 -> 0.0084 | Improved |
| use ViT encoder instead of CNN encoder | Image-to-Image model · SVIB-dSprites Medium | LPIPS: 0.1251 -> 0.0359 | Improved |
| use ViT encoder instead of CNN encoder | Image-to-Image model · SVIB-CLEVR Easy | LPIPS: 0.0593 -> 0.0583 | No clear effect |
| use ViT encoder instead of CNN encoder | Image-to-Image model · SVIB-CLEVRTex Easy | LPIPS: 0.2938 -> 0.2529 | Mixed |
| use slot-based latent (SSM-Slot) instead of single-vector VAE latent | State-space model · SVIB-dSprites Easy | LPIPS: 0.1284 -> 0.0534 | Improved |
| use slot-based latent (SSM-Slot) instead of single-vector VAE latent | State-space model · SVIB-CLEVR Hard | LPIPS: 0.1884 -> 0.1139 | Improved |
| use slot-based latent (SSM-Slot) instead of single-vector VAE latent | State-space model · SVIB-CLEVRTex Hard | LPIPS: 0.3745 -> 0.3590 | Improved |
| SSM-Slot instead of I2I-ViT | best state-space vs best image-to-image · SVIB-CLEVR Medium | LPIPS: 0.1263 -> 0.0533 | Improved |
| SSM-Slot instead of I2I-ViT | best state-space vs best image-to-image · SVIB-dSprites Hard | LPIPS: 0.2326 -> 0.1834 | Improved |
| SSM-Slot instead of I2I-ViT | best state-space vs best image-to-image · SVIB-CLEVRTex Easy | LPIPS: 0.2529 -> 0.3429 | Hurt |
| ground-truth tokens (Oracle) instead of learned perception | Oracle vs I2I-ViT · SVIB-CLEVRTex Easy | LPIPS: 0.2529 -> 0.1414 | Improved |
| ground-truth tokens (Oracle) instead of learned perception | Oracle vs SSM-Slot · SVIB-dSprites Hard | LPIPS: 0.1834 -> 0.0064 | Improved |
Closest papers by meaning
- Trimming the Long-Tail of Visual World Modeling Evaluation (NeurIPS 2026 · similarity 0.88)
- Thinking with Imagination: Agentic Visual Spatial Reasoning with World Simulators (NeurIPS 2026 · similarity 0.88)
- Dreamweaver: Learning Compositional World Models from Pixels (ICLR 2025 · similarity 0.88)
- Learning to See Before Seeing: Demystifying LLM Visual Priors from Language Pre-training (ICLR 2026 · similarity 0.88)
- Bridging the gap to real-world language-grounded visual concept learning (NeurIPS 2025 · similarity 0.88)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.87)
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.
