Abstract: The current training strategies based on knowledge distillation for image captioning assume that each learning model possesses complete learning value, lacking review and guidance mechanisms ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback