segformer_b1_cityscapes_1024x1024_160k.yml

_base_: '../_base_/cityscapes_1024x1024.yml'

batch_size: 2
iters: 160000

model:
  type: SegFormer_B1
  num_classes: 19
  pretrained: https://bj.bcebos.com/paddleseg/dygraph/mix_vision_transformer_b1.tar.gz

optimizer:
  _inherited_: False
  type: AdamW
  beta1: 0.9
  beta2: 0.999
  weight_decay: 0.01

lr_scheduler:
  type: PolynomialDecay
  learning_rate: 0.00006
  power: 1

loss:
  types:
    - type: CrossEntropyLoss
  coef: [1]

test_config:
    is_slide: True
    crop_size: [1024, 1024]
    stride: [768, 768]