LogicDiag identifies conflicts in pseudo labels using symbolic knowledge and resolves them through logic-induced diagnoses. This approach improves performance by correcting erroneous labels and reducing error accumulation.
A new model that incorporates finite memory to align both the short-term and long-term video context with the language expression in an efficient manner.
We propose GMMSeg, a new family of segmentation models that rely on a dense generative classifier. GMMSeg is the first semantic segmentation method that reports promising results on both closed-set and open-world scenarios by using a single model instance.