26 lines
622 B
Python
26 lines
622 B
Python
# model settings
|
|
model = dict(
|
|
type='ImageClassifier',
|
|
backbone=dict(
|
|
type='VisionTransformer',
|
|
arch='b',
|
|
img_size=224,
|
|
patch_size=16,
|
|
drop_rate=0.1,
|
|
init_cfg=[
|
|
dict(
|
|
type='Kaiming',
|
|
layer='Conv2d',
|
|
mode='fan_in',
|
|
nonlinearity='linear')
|
|
]),
|
|
neck=None,
|
|
head=dict(
|
|
type='VisionTransformerClsHead',
|
|
num_classes=1000,
|
|
in_channels=768,
|
|
loss=dict(
|
|
type='LabelSmoothLoss', label_smooth_val=0.1,
|
|
mode='classy_vision'),
|
|
))
|