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

Table 2

Hyperparameters set in experiments.

Parameters	Description	Performance comparison	Ablation study
	Initial learning rate	0.05	0.05
	Weight decay	1e − 4	1e-4
	Momentum	0.9	0.9
	Dimension of image view features	2048	2048
	Dimension of cross-modal embedding	1024	{512, 1024, 2048}
	Dimension of metric embedding	{64, 128}	{64, 128, 256, 512}
	Number of cross-model embedding	2	{1, 2, 3, 4}
	Layers of multimodal projector	4	4
	Layers of predictor	2	2