_base_ = './segformer_b0_coco.py' model = dict( pretrained= 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b4_20220624-d588d980.pth', backbone=dict( embed_dims=64, num_layers=[3, 8, 27, 3], ), decode_head=dict( in_channels=[64, 128, 320, 512], channels=768, ), )