[ad_1]
Contrastive studying has emerged as a transformative methodology for studying efficient visible representations via the alignment of picture and textual content embeddings. Nonetheless, pairwise similarity computation in contrastive loss between picture and textual content pairs poses computational challenges. This paper presents a novel weakly supervised pre-training of imaginative and prescient fashions on web-scale image-text information. The proposed methodology reframes pre-training on image-text information as a classification activity. Consequently, it eliminates the necessity for pairwise similarity computations in contrastive loss, attaining a exceptional 2.7…
[ad_2]
Source link
