README.md 5.81 KB
Newer Older
arnedewall's avatar
arnedewall committed
1
# Deep Learning Models for BigEarthNet-MM with 19 Classes
gencersumbul's avatar
gencersumbul committed
2
This repository contains code to use the [multi-modal BigEarthNet (BigEarthNet-MM)](http://bigearth.net) archive with the nomenclature of 19 classes for deep learning applications. The nomenclature of 19 classes was defined by interpreting and arranging the CORINE Land Cover (CLC) Level-3 nomenclature based on the properties of Sentinel-2 images. This class nomenclature is the product of a collaboration between the [Direção-Geral do Território](http://www.dgterritorio.pt/) in Lisbon, Portugal and the [Remote Sensing Image Analysis (RSiM)](https://www.rsim.tu-berlin.de/) group at TU Berlin, Germany.
arnedewall's avatar
arnedewall committed
3

gencersumbul's avatar
gencersumbul committed
4
5
6
7
8
9
10
11
12
13
14
15
16
17
If you use the BigEarthNet-MM archive or our pre-trained models, please cite the paper given below:

> G. Sumbul, A. d. Wall, T. Kreuziger, F. Marcelino, H. Costa, P. Benevides, M. Caetano, B. Demir, V. Markl, “[BigEarthNet-MM: A Large Scale Multi-Modal Multi-Label Benchmark Archive for Remote Sensing Image Classification and Retrieval](https://arxiv.org/abs/2105.07921)”, arXiv:2105.07921, 2021.

```
@misc{BigEarthNet-MM,
      title={BigEarthNet-MM: A Large Scale Multi-Modal Multi-Label Benchmark Archive for Remote Sensing Image Classification and Retrieval}, 
      author={Gencer Sumbul and Arne de Wall and Tristan Kreuziger and Filipe Marcelino and Hugo Costa and Pedro Benevides and Mário Caetano and Begüm Demir and Volker Markl},
      year={2021},
      eprint={2105.07921},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}
```
arnedewall's avatar
arnedewall committed
18
19
20
21
22
23
24
25
26
27
28
29
30

## Pre-trained Deep Learning Models
We provide code and model weights for the following deep learning models that have been pre-trained on BigEarthNet-MM with the nomenclature of 19 classes for scene classification:


| Model Names  | Pre-Trained TensorFlow Models                                | 
| ------------ | ------------------------------------------------------------ | 
| VGG16        | [VGG16.zip](http://bigearth.net/static/pretrained-models/BigEarthNet-MM_19-Classes/VGG16.zip) | 
| VGG19        | [VGG19.zip](http://bigearth.net/static/pretrained-models/BigEarthNet-MM_19-Classes/VGG19.zip) | 
| ResNet50     | [ResNet50.zip](http://bigearth.net/static/pretrained-models/BigEarthNet-MM_19-Classes/ResNet50.zip) | 
| ResNet101    | [ResNet101.zip](http://bigearth.net/static/pretrained-models/BigEarthNet-MM_19-Classes/ResNet101.zip) | 
| ResNet152    | [ResNet152.zip](http://bigearth.net/static/pretrained-models/BigEarthNet-MM_19-Classes/ResNet152.zip) |

tristan.kreuziger's avatar
tristan.kreuziger committed
31
The TensorFlow code for these models can be found [here](https://git.tu-berlin.de/rsim/bigearthnet-models-tf).
arnedewall's avatar
arnedewall committed
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46

The pre-trained models associated to other deep learning libraries will be released soon.

## Generation of Training/Test/Validation Splits
After downloading the raw images from http://bigearth.net, they need to be prepared for your ML application. We provide the script `prep_splits_19_classes.py` for this purpose. It generates consumable data files (i.e., TFRecord) for training, validation and test splits which are suitable to use with TensorFlow. Suggested splits can be found with corresponding csv files under `splits` folder. The following command line arguments for `prep_splits_19_classes.py` can be specified:

* `-r1` or `--root_folder_s1`: The root folder containing the raw images of the downloaded BigEarthNet-S1 dataset.
* `-r2` or `--root_folder_s2`: The root folder containing the raw images of the downloaded BigEarthNet-S2 dataset.
* `-o` or `--out_folder`: The output folder where the resulting files will be created.
* `-n` or `--splits`: A list of CSV files each of which contains the patch names of corresponding split.
* `-l` or `--library`: A flag to indicate for which ML library data files will be prepared: TensorFlow.
* `--update_json`: A flag to indicate that this script will also change the original json files of the BigEarthNet-MM by updating labels 

To run the script, either the GDAL or the rasterio package should be installed. The TensorFlow v1 package should also be installed. The script is tested with Python 3.6, TensorFlow 1.15, and Ubuntu 16.04. 

gencersumbul's avatar
gencersumbul committed
47
**Note**: In the experiments, we did not use the Sentinel-2 patches that are fully covered by seasonal snow, cloud, and cloud shadow. Thus, they are not included in the training, test and validation sets constructed by the provided scripts (see the list of Sentinel-2 patches with seasonal snow [here](http://bigearth.net/static/documents/patches_with_seasonal_snow.csv) and that of cloud and cloud shadow [here](http://bigearth.net/static/documents/patches_with_cloud_and_shadow.csv)). 
arnedewall's avatar
arnedewall committed
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76

## License
The BigEarthNet Archive is licensed under the **Community Data License Agreement – Permissive, Version 1.0** ([Text](https://cdla.io/permissive-1-0/)).

The code in this repository to facilitate the use of the archive is licensed under the **MIT License**:

```
MIT License

Copyright (c) 2021 The BigEarthNet Authors

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
```