22 lines
503 B
Python
22 lines
503 B
Python
|
# model settings
|
||
|
|
model = dict(
|
||
|
|
type='ImageClassifier',
|
||
|
|
backbone=dict(
|
||
|
|
type='VisionTransformer',
|
||
|
|
arch='dinov2-giant',
|
||
|
|
img_size=518,
|
||
|
|
patch_size=14,
|
||
|
|
layer_scale_init_value=1e-5,
|
||
|
|
layer_cfgs=dict(ffn_type='swiglu_fused'),
|
||
|
|
),
|
||
|
|
neck=None,
|
||
|
|
head=None)
|
||
|
|
|
||
|
|
data_preprocessor = dict(
|
||
|
|
# RGB format normalization parameters
|
||
|
|
mean=[123.675, 116.28, 103.53],
|
||
|
|
std=[58.395, 57.12, 57.375],
|
||
|
|
# convert image from BGR to RGB
|
||
|
|
to_rgb=True,
|
||
|
|
)
|