Research map / Vision-language and multimodal / Cross-modal learning
What Images Cannot Say: Language-Guided Olfactory Representation Learning
Eleftherios Tsonis, Xi WANG, Vicky Kalogeiton
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 context and inferred-smell descriptors to object labels in text supervision (ablation) | SCENT · New York Smells S2T | R@5: 8.0 -> 11.9 | Improved |
| add context and inferred-smell descriptors to object labels (ablation) | SCENT · New York Smells S2I | R@5: 22.1 -> 23.0 | Improved |
| align smell jointly with image and text instead of image only (frozen CLIP) (ablation) | SCENT · New York Smells | R@5: 12.1 -> 14.3 | Improved |
| use object-only text instead of image alignment (ablation) | SCENT · New York Smells | R@5: 12.1 -> 8.8 | Hurt |
| use native three-stream design instead of Image Bridge (ablation) | SCENT · New York Smells S2T | R@5: 9.1 -> 11.9 | Improved |
| use native three-stream design instead of Image Bridge (ablation) | SCENT · New York Smells S2IT | R@5: 21.4 -> 23.3 | Improved |
| use VLM annotations (Qwen3-VL-30B) instead of text-only LLM annotations (ablation) | SCENT · New York Smells S2T | R@20: 23.7 -> 29.2 | Improved |
| use stronger annotating VLM (Qwen3-VL-30B over Qwen2.5-VL-72B) (ablation) | SCENT · New York Smells S2T | R@5: 8.1 -> 11.9 | Improved |
| use smaller annotating VLM (Qwen3-VL-8B) (ablation) | SCENT · New York Smells S2T | R@5: 11.9 -> 6.5 | Hurt |
| use decoded synthesis from latent decomposition instead of raw recombination (ablation) | SCENT · zero-shot recombination retrieval | R@20: 9.8 -> 13.7 | Improved |
| replace NYS with SCENT on open scenes | SCENT · New York Smells open scenes S2T | R@5: 9.2 -> 13.0 | Improved |
| native text retrieval versus image bridge | SCENT · New York Smells T2S | R@5: 9.9 -> 7.2 | Hurt |
Closest papers by meaning
- Can Transformers Smell Like Humans? (NeurIPS 2024 · similarity 0.85)
- Causal Tracing of Object Representations in Large Vision Language Models: Mechanistic Interpretability and Hallucination Mitigation (AAAI 2026 · similarity 0.83)
- VCU-Bridge: Hierarchical Visual Connotation Understanding via Semantic Bridging (CVPR 2026 · similarity 0.82)
- SCENT: Robust Spatiotemporal Learning for Continuous Scientific Data via Scalable Conditioned Neural Fields (ICML 2025 · similarity 0.82)
- SNIFFER: Multimodal Large Language Model for Explainable Out-of-Context Misinformation Detection (CVPR 2024 · similarity 0.82)
- Same or Not? Enhancing Visual Perception in Vision-Language Models (CVPR 2026 · similarity 0.82)
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.
