在 Keras 中具有多个 U-Net 实现的帮助程序包以及在处理图像语义分割任务时有用的实用工具_python_代码_下载

[](https://github.com/karolzak/keras-unet/actions?query=workflow%3A%22Python+package%22) [](https://pypi.org/project/keras-unet/) [](https://pepy.tech/project/keras-unet) [](https://pepy.tech/project/keras-unet/month) [](https://github.com/karolzak/keras-unet/blob/master/LICENSE) <!--  --> **Share**: [](http://twitter.com/share?text=Check%20out%20Keras-Unet%20Python%20package%20which%20helps%20you%20to%20build%20image%20segmentation%20models%20in%20TF%20and%20Keras&url=https://github.com/karolzak/keras-unet/&hashtags=python,computervision,semanticsegmentation,unet,keras,tensorflow,deeplearning,ML,AI) [](http://www.linkedin.com/shareArticle?mini=true&url=https://github.com/karolzak/keras-unet&title=Keras%20UNet%20python%20package) # About Helper package with multiple U-Net implementations in Keras as well as useful utility tools helpful when working with image segmentation tasks # Features: - [x] U-Net models implemented in Keras - [x] Vanilla U-Net implementation based on [the original paper](https://arxiv.org/pdf/1505.04597.pdf) - [x] Customizable U-Net - [x] U-Net optimized for satellite images based on [DeepSense.AI Kaggle competition entry](https://deepsense.ai/deep-learning-for-satellite-imagery-via-image-segmentation/) - [x] Utility functions: - [x] Plotting images and masks with overlay - [x] Plotting images masks and predictions with overlay (prediction on top of original image) - [x] Plotting training history for metrics and losses - [x] Cropping smaller patches out of bigger image (e.g. satellite imagery) using sliding window technique (also with overlap if needed) - [x] Plotting smaller patches to visualize the cropped big image - [x] Reconstructing smaller patches back to a big image - [x] Data augmentation helper function - [x] Notebooks (examples): - [x] Training custom U-Net for whale tails segmentation - [ ] Semantic segmentation for satellite images - [x] Semantic segmentation for medical images [ISBI challenge 2015](https://biomedicalimaging.org/2015/program/isbi-challenges/) # Installation: ```bash pip install git+https://github.com/karolzak/keras-unet ``` or ```bash pip install keras-unet ``` # Usage examples: - U-Net implementations in Keras: - [Vanilla U-Net](#Vanilla-U-Net) - [Customizable U-Net](#Customizable-U-Net) - [U-Net for satellite images](#U-Net-for-satellite-images) - Utils: - [Plot training history](#Plot-training-history) - [Plot images and segmentation masks](#Plot-images-and-segmentation-masks) - [Get smaller patches/crops from bigger image](#Get-smaller-patches/crops-from-bigger-image) - [Plot small patches into single big image](#Plot-small-patches-into-single-big-image) - [Reconstruct a bigger image from smaller patches/crops](#Reconstruct-a-bigger-image-from-smaller-patches/crops) <br> ### Vanilla U-Net [Model scheme can be viewed here](https://raw.githubusercontent.com/karolzak/keras-unet/master/docs/vanilla_unet.png) ```python from keras_unet.models import vanilla_unet model = vanilla_unet(input_shape=(512, 512, 3)) ``` [[back to usage examples]](#usage-examples) <br> ### Customizable U-Net [Model scheme can be viewed here](https://raw.githubusercontent.com/karolzak/keras-unet/master/docs/custom_unet.png) ```python from keras_unet.models import custom_unet model = custom_unet( input_shape=(512, 512, 3), use_batch_norm=False, num_classes=1, filters=64, dropout=0.2, output_activation='sigmoid') ``` [[back to usage examples]](#usage-examples) <br> ### U-Net for satellite images [Model scheme can be viewed here](https://raw.githubusercontent.com/karolzak/keras-unet/master/docs/satellite_unet.png) ```python from keras_unet.models import satellite_unet model = satellite_unet(input_shape=(512, 512, 3)) ``` [[back to usage examples]](#usage-examples) <br> ### Plot training history ```python history = model.fit_generator(...) from keras_unet.utils import plot_segm_history plot_segm_history( history, # required - keras training history object metrics=['iou', 'val_iou'], # optional - metrics names to plot losses=['loss', 'val_loss']) # optional - loss names to plot ``` Output:   [[back to usage examples]](#usage-examples) <br> ### Plot images and segmentation masks ```python from keras_unet.utils import plot_imgs plot_imgs( org_imgs=x_val, # required - original images mask_imgs=y_val, # required - ground truth masks pred_imgs=y_pred, # optional - predicted masks nm_img_to_plot=9) # optional - number of images to plot ``` Output:  [[back to usage examples]](#usage-examples) <br> ### Get smaller patches/crops from bigger image ```python from PIL import Image import numpy as np from keras_unet.utils import get_patches x = np.array(Image.open("../docs/sat_image_1.jpg")) print("x shape: ", str(x.shape)) x_crops = get_patches( img_arr=x, # required - array of images to be cropped size=100, # default is 256 stride=100) # default is 256 print("x_crops shape: ", str(x_crops.shape)) ``` Output: ```output x shape: (1000, 1000, 3) x_crops shape: (100, 100, 100, 3) ``` [[back to usage examples]](#usage-examples) <br> ### Plot small patches into single big image ```python from keras_unet.utils import plot_patches print("x_crops shape: ", str(x_crops.shape)) plot_patches( img_arr=x_crops, # required - array of cropped out images org_img_size=(1000, 1000), # required - original size of the image stride=100) # use only if stride is different from patch size ``` Output: ```output x_crops shape: (100, 100, 100, 3) ```  [[back to usage examples]](#usage-examples) <br> ### Reconstruct a bigger image from smaller patches/crops ```python import matplotlib.pyplot as plt from keras_unet.utils import reconstruct_from_patches print("x_crops shape: ", str(x_crops.shape)) x_reconstructed = reconstruct_from_patches( img_arr=x_crops, # required - array of cropped out images org_img_size=(1000, 1000), # required - original size of the image stride=100) # use only if stride is different from patch size print("x_reconstructed shape: ", str(x_reconstructed.shape)) plt.figure(figsize=(10,10)) plt.imshow(x_reconstructed[0]) plt.show() ``` Output: ```output x_crops shape: (100, 100, 100, 3) x_reconstructed shape: (1, 1000, 1000, 3) ```  [[back to usage examples]](#usage-examples)