Gated Object-Attribute Matching Network for Detailed Image Caption

Table 1

The performance of the state-of-the-art image captioning models on the Flickr30k and MSCOCO testing splits.

Model	Flickr30k						MSCOCO
	B@1	B@2	B@3	B@4	METEOR	CIDEr	B@1	B@2	B@3	B@4	METEOR	CIDEr
DeepV-SAlign [7]	0.573	0.369	0.240	0.157	0.153	0.247	0.625	0.450	0.321	0.230	0.195	0.660
Soft-Attention [26]	0.667	0.434	0.288	0.191	0.185	—	0.707	0.492	0.344	0.243	0.239	—
Hard-Attention [26]	0.669	0.439	0.296	0.199	0.185	—	0.718	0.504	0.357	0.250	0.230	—
Attribute-FCN [15]	0.647	0.460	0.324	0.230	0.189	—	0.709	0.537	0.402	0.304	0.243	—
Adaptive-Attention [11]	0.677	0.494	0.354	0.251	0.204	0.531	0.742	0.580	0.439	0.332	0.266	1.085
Attribute-CNN + LSTM [14]	—	—	—	—	—	—	—	0.56	0.42	0.31	0.26	0.94
NBT [13]	0.720	—	—	0.285	0.231	0.575	0.759	—	—	0.349	0.274	1.089
Up-Down [12]	—	—	—	—	—	—	0.802	0.641	0.491	0.369	0.276	1.179
Ours (Box proposed)	0.711	0.507	0.393	0.266	0.211	0.630	0.753	0.592	0.462	0.341	0.266	0.954
Ours (RL)	0.735	0.522	0.401	0.297	0.219	0.674	0.772	0.620	0.476	0.352	0.270	1.098