akhilesh64 / resunet-a Goto Github PK

Implementation of the paper "ResUNet-a: a deep learning framework for semantic segmentation of remotely sensed data" in TensorFlow.

License: MIT License

Python 100.00%

semantic-segmentation segmentation resunet boundary-detection deep-learning tensorflow keras convolutional-neural-networks image-segmentation satellite-imagery

resunet-a's Introduction

ResUnet-a

This repository contains implementation of the paper "ResUNet-a: a deep learning framework for semantic segmentation of remotely sensed data" in TensorFlow for the detection of plot boundaries specifically.

Usage

Clone this repo using :

git clone https://github.com/Akhilesh64/ResUnet-a

Install the requirements using :

pip install -r requirements.txt

To start model training run the main.py file with following arguments :

python main.py --image_size 256 --batch_size 8 --num_classes 2 --validation_split 0.2 --epochs 100 --image_path ./images --gt_path ./gt --layer_norm batch --model_save_path ./ --checkpoint_mode epochs

To produce model predictions on a directory of test images run script predict.py with the following arguments :

python predict.py --image_size 256 --num_classes 2 --image_path ./test --model_path ./model.h5 --output_path ./results

Results

Original Image Groundtruth Predicted

Citation

The arvix version of the paper can found at the following link.

If you find this repo useful please cite the original authors :

@article{DIAKOGIANNIS202094,
title = "ResUNet-a: A deep learning framework for semantic segmentation of remotely sensed data",
journal = "ISPRS Journal of Photogrammetry and Remote Sensing",
volume = "162",
pages = "94 - 114",
year = "2020",
issn = "0924-2716",
doi = "https://doi.org/10.1016/j.isprsjprs.2020.01.013",
url = "http://www.sciencedirect.com/science/article/pii/S0924271620300149",
author = "Foivos I. Diakogiannis and François Waldner and Peter Caccetta and Chen Wu",
keywords = "Convolutional neural network, Loss function, Architecture, Data augmentation, Very high spatial resolution"
}

resunet-a's People

Contributors

Stargazers

Watchers

Forkers

giser18 haiderali2580 bumie-e xuejie7 westamine collaboratord renzosgit gebeyehua kcsumesh govindfromindia rao-monu tarequebashar

resunet-a's Issues

Where can i get dataset?

Hi, i have some problems with running project. What do i must place in ./images and ./gt directories?

Such images?
And another question. What i must write here
python main.py --image_size 256 --batch_size 8 --num_classes 2 --validation_split 0.2 --epochs 100 --image_path ./images --gt_path ./gt --layer_norm batch --model_save_path ./ --checkpoint_mode epochs
if my images have size 512x512? What means these parameters?
--image_size 256 --batch_size 8 --num_classes 2 --validation_split 0.2 --epochs 100

Error cloning repo

Hi, Akhilesh! When cloning the repo, a message is received: ERROR: File "setup.py" not found for legacy project git+git://github.com/Akhilesh64/ResUnet-a.git.

How to continue to train?

Hi, I am trying to use your codes on Google Colab. But sometimes Colab suddenly closes itself. Is there a way to continue training?

I didn't see any flag in main.py about this.

Getting Value Error

ValueError: The two structures don't have the same sequence length. Input structure has length 4, while shallow structure has length 3.
Getting this error after running the code. #

Want to know the python version u r using for tensorflow GPU to run this code

I am facing an error while using tensorflow-gpu==2.0.0 , It is asking for downgrading the python version.

Please let me know the python version you have used for Tensorflow gpu.

Question about distance, boundary, segmentation and watershed algorithm.

Hello, at first, i like to thank you about the code that you have shared. My question is, how can we obtain 1) the probability of belonging to a field; 2) the probability
of belonging to a boundary; and 3) the distance to the closest
boundary with the corresponding code and have you used watershed algorithm as the resunet article suggests.

`dist = ZeroPadding2D(padding=1)(x1)
dist = Conv2D(32, 3)(dist)
dist = self.normalization(dist, self.layer_norm)
dist = Activation('relu')(dist)
dist = ZeroPadding2D(padding=1)(dist)
dist = Conv2D(32, 3)(dist)
dist = self.normalization(dist, self.layer_norm)
dist = Activation('relu')(dist)
dist = Conv2D(self.num_classes, 1, activation='softmax', name = 'distance')(dist)

    bound = Concatenate(axis=-1)([x, dist])
    bound = ZeroPadding2D(padding=1)(bound)
    bound = Conv2D(32, 3)(bound)
    bound = self.normalization(bound, self.layer_norm)
    bound = Activation('relu')(bound)
    bound = Conv2D(self.num_classes, 1, activation='sigmoid', name = 'boundary')(bound)

    seg = Concatenate(axis=-1)([x,bound,dist])
    seg = ZeroPadding2D(padding=1)(seg)
    seg = Conv2D(32, 3)(seg)
    seg = self.normalization(seg, self.layer_norm)
    seg = Activation('relu')(seg)
    seg = ZeroPadding2D(padding=1)(seg)
    seg = Conv2D(32, 3)(seg)
    seg = self.normalization(seg, self.layer_norm)
    seg = Activation('relu')(seg)
    seg = Conv2D(self.num_classes, 1, activation='softmax', name = 'segmentation')(seg)

    model = Model(inputs = input, outputs={'seg': seg, 'bound': bound, 'dist': dist})`

about d7v1

Hello, I tried to conduct d7v1 model but there is error, how should I modify? thank you!

tensorflow-addons-0.6.0

How to use pip to install tensorflow-addons-0.6.0? I find not it. Thank you