通过本教程，你将快速学会PaddleSeg的API调用，轻松进行语义分割模型的训练、评估和预测。我们将以BiSeNetV2和视盘分割数据集为例一步一步的教导你如何调用API进行模型、数据集、损失函数、优化器等模块的构建。
如果对配置化调用方式更感兴趣，可参考10分钟上手PaddleSeg教程。
Note：请在AI studio上fork本项目的最新版本，然后运行使用。
Note：若想更详细地了解PaddleSeg API，请阅读API文档
PaddleSeg安装及环境配置
!pip install paddleseg
模型训练
1. 构建模型
from paddleseg.models import BiSeNetV2 model = BiSeNetV2(num_classes=2, lambd=0.25, align_corners=False, pretrained=None)
2. 构建训练集
# 构建训练用的transforms import paddleseg.transforms as T transforms = [ T.Resize(target_size=(512, 512)), T.RandomHorizontalFlip(), T.Normalize() ] # 构建训练集 from paddleseg.datasets import OpticDiscSeg train_dataset = OpticDiscSeg( dataset_root='data/optic_disc_seg', transforms=transforms, mode='train' )
3. 构建验证集
# 构建验证用的transforms import paddleseg.transforms as T transforms = [ T.Resize(target_size=(512, 512)), T.Normalize() ] # 构建验证集 from paddleseg.datasets import OpticDiscSeg val_dataset = OpticDiscSeg( dataset_root='data/optic_disc_seg', transforms=transforms, mode='val' )
4. 构建优化器
import paddle # 设置学习率 base_lr = 0.01 lr = paddle.optimizer.lr.PolynomialDecay(base_lr, power=0.9, decay_steps=1000, end_lr=0) optimizer = paddle.optimizer.Momentum(lr, parameters=model.parameters(), momentum=0.9, weight_decay=4.0e-5)
5. 构建损失函数
为了适应多路损失，损失函数应构建成包含'types'和'coef'的dict，如下所示。 其中losses['type']表示损失函数类型， losses['coef']为对应的系数。需注意len(losses['types'])应等于len(losses['coef'])。
from paddleseg.models.losses import CrossEntropyLoss losses = {} losses['types'] = [CrossEntropyLoss()] * 5 losses['coef'] = [1]* 5
6.训练
from paddleseg.core import train train( model=model, train_dataset=train_dataset, val_dataset=val_dataset, optimizer=optimizer, save_dir='output', iters=1000, batch_size=4, save_interval=200, log_iters=10, num_workers=0, losses=losses, use_vdl=True)
模型评估
1. 构建模型
from paddleseg.models import BiSeNetV2 model = BiSeNetV2(num_classes=2, lambd=0.25, align_corners=False, pretrained=None)
2. 加载模型参数
model_path = 'output/best_model/model.pdparams' if model_path: para_state_dict = paddle.load(model_path) model.set_dict(para_state_dict) print('Loaded trained params of model successfully') else: raise ValueError('The model_path is wrong: {}'.format(model_path))
3. 构建验证集
# 构建验证用的transforms import paddleseg.transforms as T transforms = [ T.Resize(target_size=(512, 512)), T.Normalize() ] # 构建验证集 from paddleseg.datasets import OpticDiscSeg val_dataset = OpticDiscSeg( dataset_root='data/optic_disc_seg', transforms=transforms, mode='val' )
4. 评估
from paddleseg.core import evaluate evaluate( model, val_dataset)
5. 多尺度+翻转评估
evaluate( model, val_dataset, aug_eval=True, scales=[0.75, 1.0, 1.25], flip_horizontal=True)
效果可视化
1. 构建模型
from paddleseg.models import BiSeNetV2 model = BiSeNetV2(num_classes=2, lambd=0.25, align_corners=False, pretrained=None)
2. 创建transform
import paddleseg.transforms as T transforms = T.Compose([ T.Resize(target_size=(512, 512)), T.RandomHorizontalFlip(), T.Normalize() ])
3. 构建待预测的图像列表
import os def get_image_list(image_path): """Get image list""" valid_suffix = [ '.JPEG', '.jpeg', '.JPG', '.jpg', '.BMP', '.bmp', '.PNG', '.png' ] image_list = [] image_dir = None if os.path.isfile(image_path): if os.path.splitext(image_path)[-1] in valid_suffix: image_list.append(image_path) elif os.path.isdir(image_path): image_dir = image_path for root, dirs, files in os.walk(image_path): for f in files: if os.path.splitext(f)[-1] in valid_suffix: image_list.append(os.path.join(root, f)) else: raise FileNotFoundError( '`--image_path` is not found. it should be an image file or a directory including images' ) if len(image_list) == 0: raise RuntimeError('There are not image file in `--image_path`') return image_list, image_dir image_path = 'data/optic_disc_seg/JPEGImages/N0010.jpg' # 也可以输入一个包含图像的目录 image_list, image_dir = get_image_list('data/optic_disc_seg/JPEGImages/N0010.jpg')
4. 预测
图片预测结果将会输出到保存路径save_dir当中。该路径下将生成2个目录，pseudo_color_prediction保存伪彩色预测结果图，可直接查看各个类别的预测效果，added_prediction保存伪彩色预测结果和原图的叠加效果图。
from paddleseg.core import predict predict( model, model_path='output/best_model/model.pdparams', transforms=transforms, image_list=image_list, image_dir=image_dir, save_dir='output/results' )
预测效果如下：
- 伪彩色预测结果
- 叠加效果
