Learning from Reliable Latent Prompts for Visual Recognition with Missing Modalities

Large-scale multimodal models (LMMs) have achieved superior performance in visual recognition by synergizing information across diverse, massive-scale paired modalities.

Large-scale multimodal models (LMMs) have achieved superior performance in visual recognition by synergizing information across diverse, massive-scale paired modalities. In real-world scenarios, however, missing-modality inputs are ubiquitous, causing models optimized for…