Table 2. Parameterization of model and training.

Model parameter	Value	Training parameter	Value	Training parameter	Value
Image size	256	Batch size	64	Warmup lr	0.0000005
Windows size	16	Epochs	300	Scheduler	Cosine
Embed dim	96	Warmup epochs	20	Decay epochs	30
Depth	2,2,6,2	Weight decay	0.05	Decay rate	0.1
Num heads	3,6,12,24	Optimizer	AdamW	Random erase mode	Pixel
Drop path rate	0.2	Base lr	0.0005	Mixup alpha	0.8