Visual-Text Reference Pretraining Model for Image Captioning

Table 2

The performance of the published state of the art and our model on the test sets of Visual Genome. In dense image captioning, the model receives a single image and generates a set of regions, each annotated with confidence and a caption.

Approach	B@4	M	R	C	S
ST [20]	11.1	17.0	34.5	139.9	31.1
UP-DOWN [45]	10.9	16.9	34.5	139.4	31.4
ASG [27]	17.6	22.1	44.7	202.4	40.6
VTR-PTM (ours)	20.5	27.8	45.3	185.9	50.9