Fully Convolutional Neural Network for Semantic Segmentation

Functions

Extract person from image
Extract background from image
Extract person with transparent background

modified codes from https://github.com/warmspringwinds/tf-image-segmentation

File Structure

.
├── data                # your testing data dir
|   └──                 # your image data goes here 
|
├── fcn_16s_checkpoint     # place pretrained fcn16 model here :
|     └── model_fcn16s_final.ckpt.data... # download 
|     └── model_fcn16s_final.ckpt.index   # https://www.dropbox.com/s/tmhblqcwqvt2zjo/fcn_16s.tar.gz?dl=0 and 
|     └── model_fcn16s_final.ckpt.meta    # unzip it here
|     
├── models              
|     └── slim
|     └──  ...
|     └──  ...
|     
├── test.py                # python script for testing trained model
|
|
├── fcn8_train.py
|
├── tf_image_segmentation
|      └── generated       # generated image files
|      └── saver           # save folder
|           └── model_fcn8s_final.ckpt.data... # download 
|           └── model_fcn8s_final.ckpt.index   # https://www.dropbox.com/s/7r6lnilgt78ljia/fcn_8s.tar.gz?dl=0
|           └── model_fcn8s_final.ckpt.meta    # unzip it here
|      |
|      └── log_folder_fcn8 # log folder
|      └── background_extracted # generated background images without person
|      └── person_extracted # generated person images without background
|     
└── README.md

Dependencies

tensorflow r.0.12 or later version
$pip3 install scikit-image Pillow matplotlib, numpy, tensorflow

Path Setting

you just have to put your working directory as an argument, stated in the usage below

--dir YOUR_DIRECTORY_TO_FCN_DEMO

Usage

$ conda create -n YOUR_ENV_NAME
$ source activate YOUR_ENV_NAME
$ cd YOUR_WORKING_DIR/FCN_demo/

# To Test segmentation of your image data, place the test data in  "~/FCN_demo/data/" directory 

$ python3 test.py --dir YOUR_DIRECTORY_TO_FCN_DEMO

ex) $ python3 test.py --dir /home/nowgeun1/Desktop/FCN_demo

# You can find the prediction of your image data in  "~/FCN_demo/tf_image_segmentation/generated/"
# You can find extracted background image in "~/FCN_demo/tf_image_segmentation/background_extracted/"
# You can find person background image in "~/FCN_demo/tf_image_segmentation/person_extracted/"

Server Usage (For Crevasse Only)

1.개발환경접속

$source activate fcn

2.모듈 파일 위치로 이동

$cd /home/nowgeun1/Desktop/FCN/

3.pretrained 된 모델 테스트

$python3 test.py

4.예측 확인

/home/nowgeun1/Desktop/FCN/tf_image_segmentation/generated/
/home/nowgeun1/Desktop/FCN/tf_image_segmentation/background_extracted/
/home/nowgeun1/Desktop/FCN/tf_image_segmentation/person_extracted/

**User Input Data 활용법

테스트 이미지 파일 저장 폴더:

YOUR_WORKING_DIR/FCN_demo/data/

ex) /home/nowgeun1/Desktop/FCN_demo/data/

예측 결과 폴더:

YOUR_WORKING_DIR/FCN_demo/tf_image_segmentation/generated/

ex) /home/nowgeun1/Desktop/FCN_demo/tf_image_segmentation/generated/

References:

http://warmspringwinds.github.io/tensorflow/tf-slim/2017/01/23/fully-convolutional-networks-(fcns)-for-image-segmentation/ https://github.com/warmspringwinds/tf-image-segmentation