Abstract:Vision-language models (VLMs) are often deployed on text-only inputs, although they are trained with images. We find that removing the vision modality causes large drops in accuracy and severe miscalibration, and the model does not behave like its original language backbone under text-only prompting. This failure is not explained only by missing semantic information. Even when text descriptions preserve key content, confidence becomes unreliable, while adding a visual signal through generated images partially restores accuracy and calibration. We propose the Latent Imagination Module (LIM), a lightweight cross-attention module that predicts imagined latent embeddings from textual input and feeds them into a frozen VLM backbone without pixel-level image synthesis. Across text-only benchmarks, unseen tasks, and missing-image scenarios, LIM improves accuracy and reduces calibration error. These results suggest that latent modality completion is a practical approach for reliable VLM inference under missing-modality.
| Comments: | 9 pages, 16 figures. Accepted at the ICLR 2026 Workshop on Principled Design for Trustworthy AI: Interpretability, Robustness, and Safety across Modalities |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2605.12517 [cs.CL] |
| (or arXiv:2605.12517v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.12517 arXiv-issued DOI via DataCite |
Submission history
From: Mingyeong Kim [view email]
[v1]
Fri, 3 Apr 2026 10:03:02 UTC (874 KB)