Abstract: Current text-only image captioning methods leverage the shared feature space of CLIP to train zero-shot image captioning using text data only, leaving feature associations and contextual ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results