Research map / Vision-language and multimodal / Visual question answering
It's Not About the Journey; It's About the Destination: Following Soft Paths Under Question-Guidance for Visual Reasoning
Monica Haurilet, Alina Roitberg, Rainer Stiefelhagen
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 question-guided graph traversal and predict from destination nodes | Visual reasoning on COG · COG test, All | accuracy: 93.7 -> 97.2 | Improved |
| Use question-guided graph traversal and predict from destination nodes | Visual reasoning on AI2D · AI2D | accuracy: 41.55 -> 43.45 | Improved |
| Use question-guided graph traversal and predict from destination nodes | Visual reasoning on CLEVR · CLEVR test, All | accuracy: 98.9 -> 97.5 | Hurt |
| Use question-guided graph traversal and predict from destination nodes | Visual reasoning on CLEVR · CLEVR test, Compare Numbers | accuracy: 96.8 -> 99.6 | Improved |
| Use question-guided graph traversal and predict from destination nodes | Visual reasoning on CLEVR · CLEVR test, Compare Attributes | accuracy: 99.1 -> 99.8 | Improved |
| Use question-guided graph traversal and predict from destination nodes | Visual reasoning on CLEVR · CLEVR test, Count | accuracy: 97.1 -> 91.3 | Hurt |
| Use question-guided graph traversal and predict from destination nodes | Visual reasoning on CLEVR · CLEVR test, Exist | accuracy: 99.3 -> 98.6 | Hurt |
| Use question-guided graph traversal and predict from destination nodes | Visual reasoning on CLEVR · CLEVR test, Query Attributes | accuracy: 99.1 -> 99.5 | Improved |
| Limit the maximal soft-path length to 1 instead of 3 (ablation) | Ours on CLEVR · CLEVR validation, Query Attributes | accuracy: 99.8% -> 53.1% | Hurt |
| Limit the maximal soft-path length to 2 instead of 3 (ablation) | Ours on CLEVR · CLEVR validation, Query Attributes | accuracy: 99.8% -> 98.1% | Hurt |
| Train on counting questions, then evaluate on unseen query-attribute questions without further training | Ours on CLEVR · CLEVR knowledge transfer, Query Attributes | accuracy: 30.6% -> 56.4% | Improved |
| Train on one question type and evaluate on a different type without further training | Ours on CLEVR · CLEVR knowledge transfer, unseen tasks | accuracy | Improved |
Closest papers by meaning
- Visual Graph Arena: Evaluating Visual Conceptualization of Vision and Multimodal Large Language Models (ICML 2025 · similarity 0.90)
- SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows (NeurIPS 2026 · similarity 0.90)
- No Labels, No Problem: Training Visual Reasoners with Multimodal Verifiers (ICLR 2026 · similarity 0.89)
- Multi-step Visual Reasoning with Visual Tokens Scaling and Verification (NeurIPS 2025 · similarity 0.89)
- Not Another Text Benchmark: Putting the “Visual" Back in Visual Question Answering for Large Video Models (NeurIPS 2026 · similarity 0.89)
- ReasonMap: Towards Fine-Grained Visual Reasoning from Transit Maps (CVPR 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.
