TAVA. TAVA: Open-Vocabulary Action Recognition via Text Retrieval Augmentation and Visual Attribute Association
