EURASIP Journal on Audio, Speech, and Music Processing

Table 1 Various hyperparameters chosen for Vi-T and Swin-T

From: Transformer-based ensemble method for multiple predominant instruments recognition in polyphonic music

Hyperparameter	Vi-T	Swin-T
Image size	72 × 72	72 × 72
Patch dimension	6 × 6	4 × 4
Hyper parameter (C)	64	96
Number of heads	8	8
Number of windows	NA	4
Number of MLP nodes	2048,1048	256, 256
Mini batch-size	256	32

Back to article page