Research map / Vision-language and multimodal / Multimodal LLMs
Show Me Examples: Inferring Visual Concepts from Image Sets
Nick Stracke, Kolja Bauer, Stefan Andreas Baumann, Miguel Angel Bautista, Joshua Susskind, Bjorn Ommer
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 |
|---|---|---|---|
| increase context set from 2 to 4 images (ablation) | Ours · ImageNet hierarchy (animal subtree) | accuracy per instantiation (%): 54.60 -> 60.21 | Improved |
| increase context set from 4 to 7 images (ablation) | Ours · ImageNet hierarchy (animal subtree) | accuracy per instantiation (%): 60.21 -> 60.87 | No clear effect |
| increase context set from 2 to 4 images (ablation) | Ours · ImageNet hierarchy (animal subtree) | diversity: 0.780 -> 0.887 | Improved |
| replace 1 of 5 context images with random (ablation) | Ours · ImageNet hierarchy (animal subtree) | accuracy per instantiation (%): 60.21 -> 51.72 | Hurt |
| replace 2 of 5 context images with random (ablation) | Ours · ImageNet hierarchy (animal subtree) | accuracy per instantiation (%): 60.21 -> 39.81 | Hurt |
| replace 2 of 5 context images with random (ablation) | Ours · ImageNet hierarchy (animal subtree) | diversity: 0.875 -> 0.943 | No clear effect |
| use proposed Set Learner model instead of Visual Prompting | Visual Prompting · ImageNet hierarchy (animal subtree) | accuracy per concept (%): 39.16 -> 55.68 | Improved |
| use proposed model instead of ILLUME+ 3B | ILLUME+ 3B · ImageNet hierarchy (animal subtree) | accuracy per concept (%): 37.15 -> 55.68 | Improved |
| use sketch queries | Ours vs Visual Prompting · Sketch queries | accuracy per concept (%): 33.23 -> 39.53 | Improved |
| use sketch context and sketch queries | Ours vs Visual Prompting · Sketch context+queries | accuracy per instantiation (%): 51.96 -> 50.37 | Hurt |
| use unseen ImageNet21k classes as context | Ours vs Visual Prompting · ImageNet21k unseen-class context | accuracy per concept (%): 35.91 -> 47.62 | Improved |
| use ILLUME+ prompt 2 instead of prompt 3 (ablation) | ILLUME+ · ImageNet hierarchy | accuracy per concept (%): 33.19 -> 37.15 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ILLUME+ 3B — ILLUME: Illuminating Your LLMs to See, Draw, and Self-Enhance (ICCV 2025) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Language-Informed Visual Concept Learning (ICLR 2024 · similarity 0.90)
- ConceptBed: Evaluating Concept Learning Abilities of Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.89)
- Revisiting the Role of Language Priors in Vision-Language Models (ICML 2024 · similarity 0.88)
- Language Model as Visual Explainer (NeurIPS 2024 · similarity 0.88)
- Bridging the gap to real-world language-grounded visual concept learning (NeurIPS 2025 · similarity 0.87)
- Bridging Vision and Language Spaces with Assignment Prediction (ICLR 2024 · 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.
