资源分类

image-segmentation-keras

2020-01-08 |

41 |

0 |

image-segmentation-keras

Image Segmentation Keras : Implementation of Segnet, FCN, UNet, PSPNet and other models in Keras.

Implementation of various Deep Image Segmentation models in keras.

Link to the full blog post with tutorial : https://divamgupta.com/image-segmentation/2019/06/06/deep-learning-semantic-segmentation-keras.html

Our Other Repositories

Attention based Language Translation in Keras
Ladder Network in Keras model achives 98% test accuracy on MNIST with just 100 labeled examples

Contributors

Divam Gupta : https://divamgupta.com

Rounaq Jhunjhunu wala

Models

Following models are supported:

model_name	Base Model	Segmentation Model
fcn_8	Vanilla CNN	FCN8
fcn_32	Vanilla CNN	FCN8
fcn_8_vgg	VGG 16	FCN8
fcn_32_vgg	VGG 16	FCN32
fcn_8_resnet50	Resnet-50	FCN32
fcn_32_resnet50	Resnet-50	FCN32
fcn_8_mobilenet	MobileNet	FCN32
fcn_32_mobilenet	MobileNet	FCN32
pspnet	Vanilla CNN	PSPNet
vgg_pspnet	VGG 16	PSPNet
resnet50_pspnet	Resnet-50	PSPNet
unet_mini	Vanilla Mini CNN	U-Net
unet	Vanilla CNN	U-Net
vgg_unet	VGG 16	U-Net
resnet50_unet	Resnet-50	U-Net
mobilenet_unet	MobileNet	U-Net
segnet	Vanilla CNN	Segnet
vgg_segnet	VGG 16	Segnet
resnet50_segnet	Resnet-50	Segnet
mobilenet_segnet	MobileNet	Segnet

Getting Started

Prerequisites

Keras 2.0
opencv for python
Theano / Tensorflow / CNTK

apt-get install -y libsm6 libxext6 libxrender-dev
pip install opencv-python

Installing

Install the module

pip install keras-segmentation

pip install git+https://github.com/divamgupta/image-segmentation-keras

or

git clone https://github.com/divamgupta/image-segmentation-kerascd image-segmentation-keras
python setup.py install

pip install will be available soon!

Pre-trained models:

from keras_segmentation.pretrained import pspnet_50_ADE_20K , pspnet_101_cityscapes, pspnet_101_voc12

model = pspnet_50_ADE_20K() # load the pretrained model trained on ADE20k datasetmodel = pspnet_101_cityscapes() # load the pretrained model trained on Cityscapes datasetmodel = pspnet_101_voc12() # load the pretrained model trained on Pascal VOC 2012 dataset# load any of the 3 pretrained modelsout = model.predict_segmentation(    inp="input_image.jpg",    out_fname="out.png")

Preparing the data for training

You need to make two folders

Images Folder - For all the training images
Annotations Folder - For the corresponding ground truth segmentation images

The filenames of the annotation images should be same as the filenames of the RGB images.

The size of the annotation image for the corresponding RGB image should be same.

For each pixel in the RGB image, the class label of that pixel in the annotation image would be the value of the blue pixel.

Example code to generate annotation images :

import cv2import numpy as np

ann_img = np.zeros((30,30,3)).astype('uint8')
ann_img[ 3 , 4 ] = 1 # this would set the label of pixel 3,4 as 1cv2.imwrite( "ann_1.png" ,ann_img )

Only use bmp or png format for the annotation images.

Download the sample prepared dataset

Download and extract the following:

https://drive.google.com/file/d/0B0d9ZiqAgFkiOHR1NTJhWVJMNEU/view?usp=sharing

You will get a folder named dataset1/

Using the python module

You can import keras_segmentation in your python script and use the API

from keras_segmentation.models.unet import vgg_unet

model = vgg_unet(n_classes=51 ,  input_height=416, input_width=608  )

model.train(    train_images =  "dataset1/images_prepped_train/",    train_annotations = "dataset1/annotations_prepped_train/",    checkpoints_path = "/tmp/vgg_unet_1" , epochs=5)

out = model.predict_segmentation(    inp="dataset1/images_prepped_test/0016E5_07965.png",    out_fname="/tmp/out.png")import matplotlib.pyplot as plt
plt.imshow(out)# evaluating the model print(model.evaluate_segmentation( inp_images_dir="dataset1/images_prepped_test/"  , annotations_dir="dataset1/annotations_prepped_test/" ) )

Usage via command line

You can also use the tool just using command line

Visualizing the prepared data

You can also visualize your prepared annotations for verification of the prepared data.

python -m keras_segmentation verify_dataset 
 --images_path="dataset1/images_prepped_train/" 
 --segs_path="dataset1/annotations_prepped_train/"  
 --n_classes=50

python -m keras_segmentation visualize_dataset 
 --images_path="dataset1/images_prepped_train/" 
 --segs_path="dataset1/annotations_prepped_train/"  
 --n_classes=50

Training the Model

To train the model run the following command:

python -m keras_segmentation train 
 --checkpoints_path="path_to_checkpoints" 
 --train_images="dataset1/images_prepped_train/" 
 --train_annotations="dataset1/annotations_prepped_train/" 
 --val_images="dataset1/images_prepped_test/" 
 --val_annotations="dataset1/annotations_prepped_test/" 
 --n_classes=50 
 --input_height=320 
 --input_width=640 
 --model_name="vgg_unet"

Choose model_name from the table above

Getting the predictions

To get the predictions of a trained model

python -m keras_segmentation predict 
 --checkpoints_path="path_to_checkpoints" 
 --input_path="dataset1/images_prepped_test/" 
 --output_path="path_to_predictions"

Model Evaluation

To get the IoU scores

python -m keras_segmentation evaluate_model 
 --checkpoints_path="path_to_checkpoints" 
 --images_path="dataset1/images_prepped_test/" 
 --segs_path="dataset1/annotations_prepped_test/"

Fine-tuning from existing segmentation model

The following example shows how to fine-tune a model with 10 classes .

from keras_segmentation.models.model_utils import transfer_weightsfrom keras_segmentation.pretrained import pspnet_50_ADE_20Kfrom keras_segmentation.models.pspnet import pspnet_50

pretrained_model = pspnet_50_ADE_20K()

new_model = pspnet_50( n_classes=51 )

transfer_weights( pretrained_model , new_model  ) # transfer weights from pre-trained model to your modelnew_model.train(    train_images =  "dataset1/images_prepped_train/",    train_annotations = "dataset1/annotations_prepped_train/",    checkpoints_path = "/tmp/vgg_unet_1" , epochs=5)