Visual-Text Reference Pretraining Model for Image Captioning

Figure 6

Changes of loss scores of different input methods during training. The abscissa is the value of the epoch, and the ordinate is the value of the loss.