Dimensions overall score 4.0

GitHub Code Pulse

No public code linked for this paper yet.

• COMPASS outperforms prior methods on the large majority of scenarios in experiments on XRF55, MM-Fi, and OctoNet under diverse single- and multiple-missing settings.result 90%
• COMPASS synthesizes target-specific proxy tokens from observed modalities using pairwise source-to-target generators in a shared latent space for missing modalities.method 90%
• Preserving a modality-complete fusion interface is a simple and effective design principle for robust multimodal sensing.technical 85%
• COMPASS ensures the fusion head always receives a fixed N-slot multimodal input with one token per modality slot.method 90%
• Missing modalities remain a major challenge for multimodal sensing because most existing methods lead to incomplete fusion and degraded cross-modal interaction.technical 85%
• COMPASS combines proxy alignment, shared-space regularization, and per-proxy discriminative supervision to make proxy tokens both representation-compatible and task-informative.method 90%
• COMPASS is built on the principle of fusion completeness where the fusion head always receives consistent input structure regardless of missing modalities.method 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card