urban_seg QQ Group: 679897018

这个项目是一个面向新手的基于遥感图片的语义分割项目。 我们使用了在4亿张图片上进行预训练的 unicom 模型，这个模型非常高效，在遥感分割任务上表现优异。 令人惊讶的是，我们仅仅使用了4张遥感图片进行训练，就能够获得非常好的效果。

This project is a beginner-friendly semantic segmentation project based on remote sensing images. We utilize the pre-trained unicom model on a dataset of 400 million images, which is highly efficient and performs exceptionally well on remote sensing segmentation tasks. Surprisingly, we achieve excellent results by training the model with just 4 remote sensing images.

如果您想快速开始，可以使用 train_one_gpu.py 来启动训练，这是个简易的代码，只有200行。 但如果您追求更好的性能，可以尝试使用稍微复杂一些的代码 train_multi_gpus.py，该代码支持多GPU训练。 请注意，train_multi_gpus.py 可能需要一些额外的配置和设置，以便正确地运行多GPU训练。确保在使用之前仔细阅读代码中的说明和文档，以确保正确设置和配置。

If you want to get started quickly, you can use train_one_gpu.py to initiate the training. It is a simple code with only 200 lines. However, if you aim for better performance, you can try using the slightly more complex train_multi_gpus.py code, which supports training on multiple GPUs. Please note that train_multi_gpus.py may require additional configurations and settings to run multi-GPU training correctly. Make sure to carefully read the instructions and documentation in the code to ensure proper setup and configuration before using it.

安装

Install

git clone https://github.com/anxiangsir/urban_seg.git
pip install -r requirements.txt

数据和预训练模型

Data and Pretrained Models

CCF卫星影像的AI分类与识别提供的数据集初赛复赛训练集，一共五张卫星遥感影像 百度云盘，密码：3ih2

The training dataset provided for the AI classification and recognition of CCF satellite images consists of a total of five satellite remote sensing images.

dataset
├── origin //5张遥感图片，有标签
├── test   //3张遥感图片，无标签，在这个任务中没有用到
└── train  //为空，通过`python preprocess.py`随机采样生成
    ├── images       
    └── labels
FP16-ViT-B-32.pt
FP16-ViT-B-16.pt
FP16-ViT-L-14.pt
FP16-ViT-L-14-336px.pt

一张GPU训练

Training on 1 GPU

1. 下载数据集到当前目录

2. 预处理数据

3. Download the dataset to the current directory.

4. Preprocess the data.

python preprocess.py

3. 训练
4. Tranining

python train_one_gpu.py

八张GPU训练

Training on 8 GPUs

1. 下载数据集到当前目录

2. 预处理数据

3. Download the dataset to the current directory.

4. Preprocess the data.

python preprocess.py

3. 训练
4. Tranining

torchrun --nproc_per_node 8 train_multi_gpus.py

和我们讨论反馈

Discuss feedback with us

QQ群：679897018 QQ Group: 679897018

引用我们

Citations

如果你觉得这个项目对你有用，欢迎引用我们的论文
If you find this project useful, please feel free to cite our paper.

@inproceedings{anxiang_2023_unicom,
  title={Unicom: Universal and Compact Representation Learning for Image Retrieval},
  author={An, Xiang and Deng, Jiankang and Yang, Kaicheng and Li, Jiawei and Feng, Ziyong and Guo, Jia and Yang, Jing and Liu, Tongliang},
  booktitle={ICLR},
  year={2023}
}