Visual-Text Reference Pretraining Model for Image Captioning

Figure 7

Changes in loss scores of different coding methods during training. SC-loss represents the training loss of VRN-SC. DC-loss represents the training loss of VRN-DC.