《Keras3：具有类似U-Net架构的图像分割》

《Keras 3 ：具有类似 U-Net 架构的图像分割》

作者：fchollet 创建日期：2019/03/20 最后修改时间：2020/04/20 描述：在 Oxford Pets 数据集上从头开始训练的图像分割模型。

（i） 此示例使用 Keras 3

在 Colab 中查看•

GitHub 源

下载数据 !!wget .robots.ox.ac.uk/~vgg/data/pets/data/images.tar.gz !!wget .robots.ox.ac.uk/~vgg/data/pets/data/annotations.tar.gz ! !curl -O thor.robots.ox.ac.uk/datasets/pets/images.tar.gz !curl -O thor.robots.ox.ac.uk/datasets/pets/annotations.tar.gz ! !tar -xf images.tar.gz !tar -xf annotations.tar.gz % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 755M 100 755M 0 0 21.3M 0 0:00:35 0:00:35 --:--:-- 22.2M % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 18.2M 100 18.2M 0 0 7977k 0 0:00:02 0:00:02 --:--:-- 7974k 准备输入图像和目标分割掩码的路径 import os input_dir = "images/" target_dir = "annotations/trimaps/" img_size = (160, 160) num_classes = 3 batch_size = 32 input_img_paths = sorted( [ os.path.join(input_dir, fname) for fname in os.listdir(input_dir) if fname.endswith(".jpg") ] ) target_img_paths = sorted( [ os.path.join(target_dir, fname) for fname in os.listdir(target_dir) if fname.endswith(".png") and not fname.startswith(".") ] ) print("Number of samples:", len(input_img_paths)) for input_path, target_path in zip(input_img_paths[:10], target_img_paths[:10]): print(input_path, "|", target_path) Number of samples: 7390 images/Abyssinian_1.jpg | annotations/trimaps/Abyssinian_1.png images/Abyssinian_10.jpg | annotations/trimaps/Abyssinian_10.png images/Abyssinian_100.jpg | annotations/trimaps/Abyssinian_100.png images/Abyssinian_101.jpg | annotations/trimaps/Abyssinian_101.png images/Abyssinian_102.jpg | annotations/trimaps/Abyssinian_102.png images/Abyssinian_103.jpg | annotations/trimaps/Abyssinian_103.png images/Abyssinian_104.jpg | annotations/trimaps/Abyssinian_104.png images/Abyssinian_105.jpg | annotations/trimaps/Abyssinian_105.png images/Abyssinian_106.jpg | annotations/trimaps/Abyssinian_106.png images/Abyssinian_107.jpg | annotations/trimaps/Abyssinian_107.png 一个输入图像和相应的分割掩码是什么样的？ from IPython.display import Image, display from keras.utils import load_img from PIL import ImageOps # Display input image #7 display(Image(filename=input_img_paths[9])) # Display auto-contrast version of corresponding target (per-pixel categories) img = ImageOps.autocontrast(load_img(target_img_paths[9])) display(img)

准备数据集以加载和矢量化批量数据 import keras import numpy as np from tensorflow import data as tf_data from tensorflow import image as tf_image from tensorflow import io as tf_io def get_dataset( batch_size, img_size, input_img_paths, target_img_paths, max_dataset_len=None, ): """Returns a TF Dataset.""" def load_img_masks(input_img_path, target_img_path): input_img = tf_io.read_file(input_img_path) input_img = tf_io.decode_png(input_img, channels=3) input_img = tf_image.