Switching to Discriminative Image Captioning by Relieving a Bottleneck of Reinforcement Learning

Ukyo Honda, Taro Watanabe, Yuji Matsumoto