English | 简体中文
AIE SDK is a deep learning training framework for remote sensing image processing. It provides local and cloud-based capabilities, rich data set acquisition, ease of use, and scalability.
# for devel
pip install -e .
# build package and install by package
bash scripts/build_pkg.sh
pip install dist/aiearth-deeplearning.tar.gz
The SDK is deeply integrated with the Alibaba Cloud AI Earth platform (referred to as the AIE platform), and can obtain sample set resources from the cloud platform, deploy models, and submit cloud-based training through the SDK. Cloud-based functions require an AIE account and a generated call token.
Please refer to the AIE Document Center for account registration.
After logging in to the AIE account in the browser, visit the token application page and generate a token according to the page prompts.
When calling aie.Authenticate() in the code, it will also prompt for a link to apply and wait for input. The effect is as follows:
Please paste the following address into the web browser, access the authorization page, and paste the personal token into the input box
https://engine-aiearth.aliyun.com/#/utility/auth-token
Personal token:
Please refer to the Quick Start Tutorial for a quick start. We also provide more tutorials to facilitate your learning and use.
For different types of tasks, we provide specialized documentation for guidance.
- Land Cover Classification
- Land Cover Classification - Semi-Supervised Learning
- Change Detection
- Building Change Detection
- Object Extraction
We also support users to customize their own algorithms based on the development paradigm provided in different Engines. For example, customizing model structure, loss functions, etc.
Task Type | Description | Model | Crop Size | mIoU | Config |
---|---|---|---|---|---|
ChangeDetection | General binary change detection | hrnet_w18 | 896x896 | N/A | config |
ChangeDetection | Building change detection | hrnet_w18 | 896x896 | N/A | config |
LandCover | General land cover classification (11 classes) | hrnet_w48 | N/A | config | |
LandCover | Semi-supervised | hrnet_w48 | N/A | config | |
TargetExtraction | Water body extraction | hrnet_w18 | N/A | config |
This project uses the Apache 2.0 open source license.
This project is maintained by the AI Earth team of Alibaba DAMO Academy. You can contact us through the following methods:
DingTalk | ||
---|---|---|
DingTalk group number: 32152986 | WeChat public account: AI Earth数知地球 | [email protected] |