Abstract: Aligned text-image encoders such as CLIP have become the de-facto model for vision-language tasks. Further-more, modality-specific encoders achieve impressive per-formances in their ...

VideoPrism is a general-purpose video encoder designed to handle a wide spectrum of video understanding tasks, including classification, retrieval, localization, captioning, and question answering. It ...

Abstract: This paper introduces a groundbreaking enhancement to image captioning through a unique approach that harnesses the combined power of the Vision Encoder-Decoder model. By leveraging the Swin ...

The existing custcat labels are reasonably balanced (C=281 is the largest, B=217 the smallest), and boxplots show that no single variable cleanly separates the four categories — distributions overlap ...