Research map / Vision-language and multimodal / Human and embodied modeling
Im-Promptu: In-Context Composition from Image Prompts
Bhishma Dedhia, Michael Chang, Jake Snell, Thomas L. Griffiths, Niraj Jha
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 object-centric slots with CET cross-attention instead of monolithic image vectors | Im-Promptu · 3D Shapes, BitMoji Faces, and CLEVr Objects composite tasks | MSE | Improved |
| Use object-centric slots with CET cross-attention instead of monolithic image vectors | Im-Promptu · 3D Shapes, BitMoji Faces, and CLEVr Objects composite tasks | FID | Improved |
| Use the pixel-space rule D̂ = C + (B − A) | Pixel baseline · 3D Shapes primitive extrapolation | MSE | Hurt |
| Use the pixel-space rule D̂ = C + (B − A) | Pixel baseline · BitMoji Faces primitive extrapolation | MSE | Hurt |
| Use a monolithic vector representation rather than object-centric slots | Im-Promptu · 3D Shapes and BitMoji Faces primitive extrapolation | MSE | Improved |
| Use a monolithic vector representation rather than object-centric slots | Im-Promptu · CLEVr Objects primitive extrapolation | FID | Hurt |
| Use a monolithic vector representation rather than object-centric slots | Im-Promptu · 3D Shapes, BitMoji Faces, and CLEVr Objects composite tasks | MSE | Hurt |
| Remove Slot Attention and represent images as discrete 4 × 4 patch latents (ablation) | Im-Promptu Patch Learner · CLEVr Objects | MSE | Improved |
| Remove Slot Attention and represent images as discrete 4 × 4 patch latents (ablation) | Im-Promptu Patch Learner · BitMoji Faces | MSE | Hurt |
| Replace CET cross-attention with concatenated slot sequences and position embeddings (ablation) | Im-Promptu Sequential Prompter · CLEVr Objects | sample efficiency | Hurt |
| Replace CET cross-attention with concatenated slot sequences and position embeddings (ablation) | Im-Promptu Sequential Prompter · 3D Shapes, BitMoji Faces, and CLEVr Objects | FID | No clear effect |
| Add more context examples | Patch Learner and Sequential Prompter · Primitive task extrapolation | MSE | Improved |
Closest papers by meaning
- Embracing Collaboration Over Competition: Condensing Multiple Prompts for Visual In-Context Learning (CVPR 2025 · similarity 0.89)
- Iterated Learning Improves Compositionality in Large Vision-Language Models (CVPR 2024 · similarity 0.89)
- Unveiling the Compositional Ability Gap in Vision-Language Reasoning Model (NeurIPS 2025 · similarity 0.88)
- Beyond Bag-of-Words: Diagnosing Compositional Binding Failures in Vision-Language Models (NeurIPS 2026 · similarity 0.88)
- Learning to Compose: Improving Object Centric Learning by Injecting Compositionality (ICLR 2024 · similarity 0.88)
- In-Context Learning Can Help Vision Language Models Overcome Training Prior (NeurIPS 2026 · similarity 0.88)
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.
