Latent Space Guided Scenario Sampling for Multimodal Segmentation Under Missing Modalities
The paper presents a novel training strategy for multimodal semantic segmentation that addresses the issue of missing modalities in remote sensing applications. By learning a scenario sampling distribution from a pretrained latent space, the method improves fine-tuning by focusing on more informative modality availability scenarios. Experimental results demonstrate that this approach outperforms traditional fine-tuning methods across various image sets and backbones.
- ▪The proposed method guides fine-tuning toward more informative modality availability scenarios.
- ▪It quantifies the effect of each scenario based on the distortion it induces in the shared latent representation.
- ▪The strategy was evaluated on three remote sensing image sets and showed improved performance over standard fine-tuning.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Computer Vision and Pattern Recognition arXiv:2605.20372 (cs) [Submitted on 19 May 2026] Title:Latent Space Guided Scenario Sampling for Multimodal Segmentation Under Missing Modalities Authors:Irem Ulku, Ö. Özgür Tanrıöver, Erdem Akagündüz View a PDF of the paper titled Latent Space Guided Scenario Sampling for Multimodal Segmentation Under Missing Modalities, by Irem Ulku and 2 other authors View PDF HTML (experimental) Abstract:Multimodal semantic segmentation benefits remote sensing analysis by combining complementary information from different sensor modalities. In real-world remote sensing applications, one or more modalities may be unavailable due to sensor failures, adverse atmospheric conditions, or data acquisition problems.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
