|
_base_ = ['./segformer_mit-b0_8xb1-160k_cityscapes-1024x1024.py']
|
|
|
|
model = dict(
|
|
backbone=dict(
|
|
init_cfg=dict(type='Pretrained', checkpoint='pretrain/mit_b2.pth'),
|
|
embed_dims=64,
|
|
num_layers=[3, 4, 6, 3]),
|
|
decode_head=dict(in_channels=[64, 128, 320, 512]))
|