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

Table 4

Accuracy on Flickr 30k. and 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	0.551	0.689	0.780	0.847	0.290	0.186
Triplet	0.530	0.664	0.753	0.823	0.281	0.164
SoftTriple	0.559	0.683	0.781	0.842	0.296	0.190
Label Relaxation	0.514	0.663	0.748	0.831	0.257	0.162
MemVir	0.543	0.691	0.769	0.849	0.287	0.189
SimSiam	0.403	0.502	0.692	0.819	0.203	0.113
SimSiam + proj	0.433	0.549	0.714	0.791	0.164	0.108
Ours ()	0.532	0.667	0.758	0.833	0.290	0.193
Ours ()	0.546	0.669	0.772	0.854	0.263	0.181