实验室4篇论文被ACM MM 2026录用

TL;DR - A research lab announced four ACM MM 2026 acceptances spanning LLM forecasting, road-anomaly segmentation, multimodal recommendation, and interior-design reasoning.

• F-LLM uses feedback correction and Lipschitz regularization to limit long-horizon forecasting errors.
• SHIELD improves detection of small road hazards, reaching 84.10% AP.
• D3ER dynamically combines shared and modality-specific features for recommendation.
• DART-I injects spatial and aesthetic priors into frozen multimodal LLMs without fine-tuning.