PTF-SimCM: A Simple Contrastive Model with Polysemous Text Fusion for Visual Similarity Metric

Table 3

Accuracy on MSCOCO. and in PTF-SimCM denote the number of embeddings and dimensions of metric embedding, respectively.

Methods	Recall@k				R-Precision	MAP@R
	k = 1	k = 2	k = 4	k = 8
Siamese	0.571	0.679	0.776	0.834	0.296	0.187
Triplet	0.557	0.656	0.762	0.823	0.286	0.172
SoftTriple	0.583	0.686	0.789	0.843	0.307	0.190
Label Relaxation	0.532	0.657	0.745	0.829	0.265	0.164
MemVir	0.579	0.693	0.794	0.840	0.296	0.192
SimSiam	0.416	0.564	0.692	0.791	0.216	0.117
SimSiam + proj	0.461	0.585	0.681	0.775	0.175	0.112
Ours ()	0.573	0.689	0.793	0.832	0.298	0.202
Ours ()	0.583	0.702	0.801	0.846	0.261	0.182