Tag
MemLearner proposes a learning-based adaptive context query method using query tokens to improve scene consistency and memory in video world models, particularly for long sequences with occlusions and dynamic objects.