PaddleSeg：使用Transformer模型对航空遥感图像分割

# clone PaddleSeg的项目!git clone https://gitee.com/paddlepaddle/PaddleSeg

#解压数据集，有两个数据集，本项目使用UDD6!mkdir work/UDD6
!unzip -oq data/data75675/UDD6.zip -d work/UDD6/

# 对数据进行crop，具体细节可查看process_data.py的代码%cd work/
!python process_data.py --tag val #处理验证集!python process_data.py --tag train #处理训练集

# 安装依赖%cd /home/aistudio/PaddleSeg
!pip install  -r requirements.txt

# 训练数据集txt生成!python tools/split_dataset_list.py \
        ../work/UDD6 train_sub train_labels_sub \
        --split 1.0 0.0 0.0 \
        --format JPG png \
        --label_class Other Facade Road Vegetation Vehicle Roof

!mv ../work/UDD6/train.txt ../work/UDD6/train_true.txt # 修改文件名

# 验证数据集txt生成!python tools/split_dataset_list.py \
        ../work/UDD6 val_sub val_labels_sub \
        --split 0.0 1.0 0.0 \
        --format JPG png \
        --label_class Other Facade Road Vegetation Vehicle Roof

!rm ../work/UDD6/train.txt #删除第二次运行生成的train.txt!mv ../work/UDD6/train_true.txt ../work/UDD6/train.txt # 将文件名改回来

# 新建.yml文件，然后将以下信息写入#!touch configs/segformer_b3_UDD.yml# 也可以直接复制准备好的文件放入该文件夹下!cp ../work/segformer_b3_UDD.yml configs/

batch_size: 2iters: 40000train_dataset:
  type: Dataset
  dataset_root: ../work/UDD6/
  train_path: ../work/UDD6/train.txt
  num_classes: 6
  transforms:
    - type: ResizeStepScaling
      min_scale_factor: 0.5
      max_scale_factor: 2.0
      scale_step_size: 0.25
    - type: RandomPaddingCrop
      crop_size: [1024, 1024]    - type: RandomHorizontalFlip
    - type: RandomDistort
      brightness_range: 0.4
      contrast_range: 0.4
      saturation_range: 0.4
    - type: Normalize
  mode: trainval_dataset:
  type: Dataset
  dataset_root: ../work/UDD6/
  val_path: ../work/UDD6/val.txt
  num_classes: 6
  transforms:
    - type: Normalize
  mode: valmodel:
  type: SegFormer_B3
  num_classes: 6
  pretrained: https://bj.bcebos.com/paddleseg/dygraph/mix_vision_transformer_b3.tar.gzoptimizer:
    type: sgdlr_scheduler:
  type: PolynomialDecay
  learning_rate: 0.001
  power: 0.9loss:
  types:
    - type: CrossEntropyLoss
  coef: [1]

# 开始训练，模型权重保存在output文件夹中，output/best_model 文件夹中为性能最好的模型!export CUDA_VISIBLE_DEVICES=0 # 设置1张可用的卡# windows下请执行以下命令# set CUDA_VISIBLE_DEVICES=0!python train.py \
       --config configs/segformer_b3_UDD.yml \
       --do_eval \
       --use_vdl \
       --save_interval 500 \
       --save_dir output

!python val.py \
       --config configs/segformer_b3_UDD.yml \
       --model_path output/best_model/model.pdparams \

!mv ../work/UDD6/val ../work/

!python predict.py \
       --config ../work/segformer_b3_UDD.yml \
       --model_path ../work/best_model/model.pdparams \
       --image_path ../work/val/src \
       --save_dir ../work/val/result \
       --is_slide \
       --crop_size 1024 1024 \
       --stride 512 512

# 展示预测的结果import matplotlib.pyplot as pltfrom PIL import Image

output = Image.open(r"../work/val/result/added_prediction/000061.JPG")

plt.figure(figsize=(20,24))#设置窗口大小plt.imshow(output), plt.axis('off')