sshuair / torchsat Goto Github PK

🔥TorchSat 🌏 is an open-source deep learning framework for satellite imagery analysis based on PyTorch.

Home Page: https://torchsat.readthedocs.io

License: MIT License

Python 99.68% Dockerfile 0.32%

pytorch deep-learning satellite satellite-imagery torchvision remote-sensing data-augmentation classification semantic-segmentation

torchsat's Introduction

TorchSat is an open-source deep learning framework for satellite imagery analysis based on PyTorch.

This project is still work in progress. If you want to know the latest progress, please check the develop branch.

Hightlight

😉 Support multi-channels(> 3 channels, e.g. 8 channels) images and TIFF file as input.
😋 Convenient data augmentation method for classification, sementic segmentation and object detection.
😍 Lots of models for satellite vision tasks, such as ResNet, DenseNet, UNet, PSPNet, SSD, FasterRCNN ...
😃 Lots of common satellite datasets loader.
😮 Training script for common satellite vision tasks.

Install

source: python3 setup.py install

How to use

Introduction
Image Classification Tutorial: Docs, Google Colab
Semantic Segmentation Tutorial: Docs
Data Augumentation: Docs, Google Colab

Features

Data augmentation

We suppose all the input images, masks and bbox should be NumPy ndarray. The data shape should be [height, width] or [height, width, channels].

pixel level

Pixel-level transforms only change the input image and will leave any additional targets such as masks, bounding boxes unchanged. It support all channel images. Some transforms only support specific input channles.

Transform	Image	masks	BBoxes
ToTensor	✓	✓	✓
Normalize	✓	✓	✓
ToGray	✓	✓	✓
GaussianBlur	✓	✓	✓
RandomNoise	✓	✓	✓
RandomBrightness	✓	✓	✓
RandomContrast	✓	✓	✓

spatial-level

Spatial-level transforms will simultaneously change both an input image as well as additional targets such as masks, bounding boxes. It support all channel images.

Transform	Image	masks	BBoxes
Resize	✓	✓	✓
Pad	✓	✓	✓
RandomHorizontalFlip	✓	✓	✓
RandomVerticalFlip	✓	✓	✓
RandomFlip	✓	✓	✓
CenterCrop	✓	✓	✓
RandomCrop	✓	✓	✓
RandomResizedCrop	✓	✓	✓
ElasticTransform	✓	✓
RandomRotation	✓	✓	✓
RandomShift	✓	✓	✓

Models

Classification

All models support multi-channels as input (e.g. 8 channels).

VGG: vgg11, vgg11_bn, vgg13, vgg13_bn, vgg16, vgg16_bn, vgg19_bn, vgg19
ResNet: resnet18, resnet34, resnet50, resnet101, resnet152, resnext50_32x4d,resnext101_32x8d, wide_resnet50_2, wide_resnet101_2
DenseNet: densenet121, densenet169, densenet201
Inception: inception_v3
MobileNet: mobilenet_v2
EfficientNet: efficientnet_b0, efficientnet_b1, efficientnet_b2, efficientnet_b3,efficientnet_b4, efficientnet_b5, efficientnet_b6, efficientnet_b7
ResNeSt: resnest50, resnest101, resnest200, resnest269

Sementic Segmentation

UNet: unet, unet34, unet101, unet152 (with resnet as backbone.)

Dataloader

Classification

Showcase

If you extend this repository or build projects that use it, we'd love to hear from you.

Reference

torchvision

Note

If you are looking for the torchvision-enhance, please checkout the enhance branch. But it was deprecated.

torchsat's People

Contributors

Stargazers

Watchers

torchsat's Issues

Is this project dead?

Hi,

Are there any plans for continuing development on this project or is it dead?

Thanks.

UNet input size dimension does not match error

code

import torch
from torchsat.models.segmentation import unet34

model = unet34(2)
inputs = torch.randn((1,3,800,800))
outputs = model(inputs)

error message:

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-7-62f9f3e6e8e0> in <module>
      1 model = unet34(2)
      2 inputs = torch.randn((1,3,800,800))
----> 3 outputs = model(inputs)

/usr/local/lib/python3.6/site-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
    491             result = self._slow_forward(*input, **kwargs)
    492         else:
--> 493             result = self.forward(*input, **kwargs)
    494         for hook in self._forward_hooks.values():
    495             hook_result = hook(self, input, result)

/usr/local/lib/python3.6/site-packages/torchsat-0.0.1-py3.6.egg/torchsat/models/segmentation/unet.py in forward(self, x)
    172         center = self.center(pool)
    173 
--> 174         dec5 = self.dec5(torch.cat([center, conv5], 1))
    175 
    176         dec4 = self.dec4(torch.cat([dec5, conv4], 1))

RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 1. Got 24 and 25 in dimension 2 at ../aten/src/TH/generic/THTensor.cpp:711
``

Can I contribute to Retina-Net?

您好！关于该库读取tif的问题

首先非常感谢您为深度学习在遥感图像的应用做的贡献。
其次，我在拜读您的源码：torchvision-enhance/torchvision_x/transforms/functional.py 时
发现有关 tifffile 的相关API还没有出现在源码当中，我印象里仅仅凭cv2好像是无法解决遥感影像的多通道问题。所以想询问一下该库现在是否可以读取tif的数据（如“sample-data/MUL_AOI_4_Shanghai_img1920.tif”）

感谢！

where can get the dataset in the document example

where can get the example data.

About noise

It's a good answers about your to-do.
https://stackoverflow.com/questions/22937589/how-to-add-noise-gaussian-salt-and-pepper-etc-to-image-in-python-with-opencv

and this,
https://gist.github.com/oeway/2e3b989e0343f0884388ed7ed82eb3b0

pretrained models?

Dear @sshuair , thanks for this great work :)

do you have any pre-trained models for segmentation and detection?

Thanks!

Best regards

Spelling

On the first line of installation.rst, Installation is misspelled. Also the exsamples directory should be examples. :)

What does "target" mean in Seg*.call(self,img,target)?

For example:

class SegResize(object):
def __ init __(self, size):
if isinstance(size, int):
self.size = (size, size)
else:
self.size = size
def __ call __(self, img, target):
return F.resize(img, self.size), F.resize(target, self.size, Image.NEAREST)

Hello, I was wondering if I could pass the "img" parameter only? Because in the F.resize function, it checks if target is None. So, it will throw an exception if I pass the "target" parameter by None.

Thanks for your sharing anyway.

what does 'ts' mean?

$ cd  tests/classification/      
$ ts make-mask-cls --filepath T51RTQ_20200513T023551_TCI.tif --width 128 --height 128 --outpath ./patched

newbie, found this command in https://torchsat.readthedocs.io/en/latest/tutorials/image-classification.html , however there is no ts or make-mask-cls command?

sshuair / torchsat Goto Github PK

torchsat's Introduction

Install

How to use

Features

Data augmentation

pixel level

spatial-level

Models

Classification

Sementic Segmentation

Dataloader

Classification

Showcase

Reference

Note

torchsat's People

Contributors

Stargazers

Watchers

Forkers

torchsat's Issues

Recommend Projects

Recommend Topics

Recommend Org