AGarioud commited on
Commit
77f46b1
1 Parent(s): 66e5f3a

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +164 -94
README.md CHANGED
@@ -7,122 +7,192 @@ size_categories:
7
  - 10B<n<100B
8
  ---
9
 
10
-
11
-
12
- # Challenge FLAIR #2: textural and temporal information for semantic segmentation from multi-source optical imagery
13
-
14
- Code : [![license](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://github.com/IGNF/FLAIR-1-AI-Challenge/blob/master/LICENSE) &emsp; Dataset : [![license](https://img.shields.io/badge/License-IO%202.0-green.svg)](https://github.com/etalab/licence-ouverte/blob/master/open-licence.md)
15
-
16
- Participate in obtaining more accurate maps for a more comprehensive description and a better understanding of our environment! Come push the limits of state-of-the-art semantic segmentation approaches on a large and challenging dataset. Get in touch at ai-challenge@ign.fr
17
-
18
-
19
- ![Alt bandeau FLAIR-IGN](images/flair_bandeau.jpg?raw=true)
20
- <img width="100%" src="images/flair-2_logos-hd.png">
21
-
22
- <div style="border-width:1px; border-style:solid; border-color:#d2db8c; padding-left: 1em; padding-right: 1em; ">
23
-
24
- <h2 style="margin-top:5px;">Links</h2>
25
-
26
-
27
- - **Datapaper : https://arxiv.org/pdf/2305.14467.pdf**
28
-
29
- - **Dataset links :** https://ignf.github.io/FLAIR/#FLAIR2
30
-
31
- - **Challenge page : https://codalab.lisn.upsaclay.fr/competitions/13447**
32
-
33
- </div>
34
- <br><br>
35
-
36
-
37
 
38
 
39
 
40
  ## Context & Data
41
 
42
- The FLAIR #2 dataset is sampled countrywide and is composed of over 20 billion annotated pixels of very high resolution aerial imagery at 0.2 m spatial resolution, acquired over three years and different months (spatio-temporal domains). Aerial imagery patches consist of 5 channels (RVB-Near Infrared-Elevation) and have corresponding annotation (with 19 semantic classes or 13 for the baselines). Furthermore, to integrate broader spatial context and temporal information, high resolution Sentinel-2 1-year time series with 10 spectral band are also provided. More than 50,000 Sentinel-2 acquisitions with 10 m spatial resolution are available.
 
 
 
43
  <br>
44
 
45
- The dataset covers 50 spatial domains, encompassing 916 areas spanning 817 km². With 13 semantic classes (plus 6 not used in this challenge), this dataset provides a robust foundation for advancing land cover mapping techniques.<br><br>
46
-
47
- <center>
48
- <table style="width:80%;max-width:700px;height:200px">
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
49
  <thead>
50
- <tr><th width=7% height=></th><th>Class</th><th style='text-align: center' width=15%>Value</th><th style='text-align: center'>Freq.-train (%)</th><th style='text-align: center'>Freq.-test (%)</th></tr>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
51
  </thead>
52
  <tbody>
53
- <tr><td bgcolor='#db0e9a'></td><td>building</td><td style='text-align: center'>1</td><td style='text-align: center'>8.14</td><td style='text-align: center'>3.26</td></tr>
54
-
55
- <tr><td bgcolor='#938e7b'></td><td>pervious surface</td><td style='text-align: center'>2</td><td style='text-align: center'>8.25</td><td style='text-align: center'>3.82</td></tr>
56
-
57
- <tr><td bgcolor='#f80c00'></td><td>impervious surface</td><td style='text-align: center'>3</td><td style='text-align: center'>13.72</td><td style='text-align: center'>5.87</td></tr>
58
-
59
- <tr><td bgcolor='#a97101'></td><td>bare soil</td><td style='text-align: center'>4</td><td style='text-align: center'>3.47</td><td style='text-align: center'>1.6</td></tr>
60
-
61
- <tr><td bgcolor='#1553ae'></td><td>water</td><td style='text-align: center'>5</td><td style='text-align: center'>4.88</td><td style='text-align: center'>3.17</td></tr>
62
-
63
- <tr><td bgcolor='#194a26'></td><td>coniferous</td><td style='text-align: center'>6</td><td style='text-align: center'>2.74</td><td style='text-align: center'>10.24</td></tr>
64
-
65
- <tr><td bgcolor='#46e483'></td><td>deciduous</td><td style='text-align: center'>7</td><td style='text-align: center'>15.38</td><td style='text-align: center'>24.79</td></tr>
66
-
67
- <tr><td bgcolor='#f3a60d'></td><td>brushwood</td><td style='text-align: center'>8</td><td style='text-align: center'>6.95</td><td style='text-align: center'>3.81</td></tr>
68
-
69
- <tr><td bgcolor='#660082'></td><td>vineyard</td><td style='text-align: center'>9</td><td style='text-align: center'>3.13</td><td style='text-align: center'>2.55</td></tr>
70
-
71
- <tr><td bgcolor='#55ff00'></td><td>herbaceous vegetation</td><td style='text-align: center'>10</td><td style='text-align: center'>17.84</td><td style='text-align: center'>19.76</td></tr>
72
-
73
- <tr><td bgcolor='#fff30d'></td><td>agricultural land</td><td style='text-align: center'>11</td><td style='text-align: center'>10.98</td><td style='text-align: center'>18.19</td></tr>
74
-
75
- <tr><td bgcolor='#e4df7c'></td><td>plowed land</td><td style='text-align: center'>12</td><td style='text-align: center'>3.88</td><td style='text-align: center'>1.81</td></tr>
76
-
77
- <tr><td bgcolor='#3de6eb'></td><td>swimming pool</td><td style='text-align: center'>13</td><td style='text-align: center'>0.01</td><td style='text-align: center'>0.02</td></tr>
78
-
79
- <tr><td bgcolor='#ffffff'></td><td>snow</td><td style='text-align: center'>14</td><td style='text-align: center'>0.15</td><td style='text-align: center'>-</td></tr>
80
-
81
- <tr><td bgcolor='#8ab3a0'></td><td>clear cut</td><td style='text-align: center'>15</td><td style='text-align: center'>0.15</td><td style='text-align: center'>0.82</td></tr>
82
-
83
- <tr><td bgcolor='#6b714f'></td><td>mixed</td><td style='text-align: center'>16</td><td style='text-align: center'>0.05</td><td style='text-align: center'>0.12</td></tr>
84
-
85
- <tr><td bgcolor='#c5dc42'></td><td>ligneous</td><td style='text-align: center'>17</td><td style='text-align: center'>0.01</td><td style='text-align: center'>-</td></tr>
86
-
87
- <tr><td bgcolor='#9999ff'></td><td>greenhouse</td><td style='text-align: center'>18</td><td style='text-align: center'>0.12</td><td style='text-align: center'>0.15</td></tr>
88
-
89
- <tr><td bgcolor='#000000'></td><td>other</td><td style='text-align: center'>19</td><td style='text-align: center'>0.14</td><td style='text-align: center'>0.04</td></tr>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
90
  </tbody>
91
  </table>
92
- </center>
93
-
94
-
95
  <br><br>
96
 
 
97
 
98
- ## Usage
 
99
 
100
- The `flair-2-config.yml` file controls paths, hyperparameters and computing ressources. The file `requirement.txt` is listing used libraries for the baselines.
 
 
101
 
102
- To launch a training/inference/metrics computation, you can either use :
103
-
104
- - ```
105
- main.py --config_file=flair-2-config.yml
106
- ```
107
-
108
- - use the `./notebook/flair-2-notebook.ipynb` notebook guiding you through data visualization, training and testing steps.
109
-
110
- A toy dataset (reduced size) is available to check that your installation and the information in the configuration file are correct.
111
-
112
- <br><br>
113
 
114
 
115
  ## Reference
116
- Please include a citation to the following article if you use the FLAIR #2 dataset:
117
 
118
  ```
119
- @article{ign2023flair2,
120
- doi = {10.13140/RG.2.2.30938.93128/2},
121
- url = {https://arxiv.org/pdf/2305.14467.pdf},
122
- author = {Garioud, Anatol and {DE Wit}, Apolline and Poupée, Marc and Valette, Marion and Giordano, Sébastien and Wattrelos, Boris},
123
- title = {FLAIR #2: textural and temporal information for semantic segmentation from multi-source optical imagery},
124
- publisher = {arXiv},
125
- year = {2023}
126
  }
127
  ```
128
 
 
7
  - 10B<n<100B
8
  ---
9
 
10
+ # Datset Card for FLAIR land-cover semantic segmentation
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11
 
12
 
13
 
14
  ## Context & Data
15
 
16
+ The hereby FLAIR (#2) dataset is sampled countrywide and is composed of over 20 billion annotated pixels of very high resolution aerial imagery at 0.2 m spatial resolution, acquired over three years and different months (spatio-temporal domains).
17
+ Aerial imagery patches consist of 5 channels (RVB-Near Infrared-Elevation) and have corresponding annotation (with 19 semantic classes or 13 for the baselines).
18
+ Furthermore, to integrate broader spatial context and temporal information, high resolution Sentinel-2 1-year time series with 10 spectral band are also provided.
19
+ More than 50,000 Sentinel-2 acquisitions with 10 m spatial resolution are available.
20
  <br>
21
 
22
+ The dataset covers 50 spatial domains, encompassing 916 areas spanning 817 km². This dataset provides a robust foundation for advancing land cover mapping techniques.<br><br>
23
+ <style type="text/css">
24
+ .tg {border-collapse:collapse;border-spacing:0;}
25
+ .tg td{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
26
+ overflow:hidden;padding:10px 5px;word-break:normal;}
27
+ .tg th{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
28
+ font-weight:normal;overflow:hidden;padding:10px 5px;word-break:normal;}
29
+ .tg .tg-kors{background-color:#3de6eb;border-color:#ffffff;text-align:left;vertical-align:top}
30
+ .tg .tg-km2t{border-color:#ffffff;font-weight:bold;text-align:left;vertical-align:top}
31
+ .tg .tg-oe15{background-color:#ffffff;border-color:#ffffff;text-align:left;vertical-align:top}
32
+ .tg .tg-r3rw{background-color:#a97101;border-color:#ffffff;text-align:left;vertical-align:top}
33
+ .tg .tg-0u95{background-color:#55ff00;border-color:#ffffff;text-align:left;vertical-align:top}
34
+ .tg .tg-zv4m{border-color:#ffffff;text-align:left;vertical-align:top}
35
+ .tg .tg-9efv{background-color:#938e7b;border-color:#ffffff;text-align:left;vertical-align:top}
36
+ .tg .tg-pop6{background-color:#fff30d;border-color:#ffffff;text-align:left;vertical-align:top}
37
+ .tg .tg-8jgo{border-color:#ffffff;text-align:center;vertical-align:top}
38
+ .tg .tg-j3z6{background-color:#194a26;border-color:#ffffff;text-align:left;vertical-align:top}
39
+ .tg .tg-oedl{background-color:#000000;border-color:#ffffff;text-align:left;vertical-align:top}
40
+ .tg .tg-40e0{background-color:#c5dc42;border-color:#ffffff;text-align:left;vertical-align:top}
41
+ .tg .tg-9xgv{background-color:#1553ae;border-color:#ffffff;text-align:left;vertical-align:top}
42
+ .tg .tg-7f0h{background-color:#6b714f;border-color:#ffffff;text-align:left;vertical-align:top}
43
+ .tg .tg-3m6m{background-color:#f80c00;border-color:#ffffff;text-align:left;vertical-align:top}
44
+ .tg .tg-2e1p{background-color:#db0e9a;border-color:#ffffff;color:#db0e9a;text-align:left;vertical-align:top}
45
+ .tg .tg-edjf{background-color:#46e483;border-color:#ffffff;text-align:left;vertical-align:top}
46
+ .tg .tg-3chm{background-color:#e4df7c;border-color:#ffffff;text-align:left;vertical-align:top}
47
+ .tg .tg-jmwx{background-color:#f3a60d;border-color:#ffffff;text-align:left;vertical-align:top}
48
+ .tg .tg-qwc7{background-color:#9999ff;border-color:#ffffff;text-align:left;vertical-align:top}
49
+ .tg .tg-69kt{background-color:#660082;border-color:#ffffff;text-align:left;vertical-align:top}
50
+ .tg .tg-x5zi{background-color:#8ab3a0;border-color:#ffffff;text-align:left;vertical-align:top}
51
+ </style>
52
+ <table class="tg">
53
  <thead>
54
+ <tr>
55
+ <th class="tg-zv4m"></th>
56
+ <th class="tg-zv4m">Class</th>
57
+ <th class="tg-8jgo">Freq.-train(%)</th>
58
+ <th class="tg-8jgo">Freq.-test(%)</th>
59
+ <th class="tg-zv4m"></th>
60
+ <th class="tg-zv4m">Class</th>
61
+ <th class="tg-8jgo">Freq.-train(%)</th>
62
+ <th class="tg-8jgo">Freq.-test(%)</th>
63
+ <th class="tg-zv4m"></th>
64
+ <th class="tg-zv4m">Class</th>
65
+ <th class="tg-8jgo">Freq.-train(%)</th>
66
+ <th class="tg-8jgo">Freq.-test(%)</th>
67
+ <th class="tg-zv4m"></th>
68
+ <th class="tg-zv4m">Class</th>
69
+ <th class="tg-8jgo">Freq.-train(%)</th>
70
+ <th class="tg-8jgo">Freq.-test(%)</th>
71
+ </tr>
72
  </thead>
73
  <tbody>
74
+ <tr>
75
+ <td class="tg-2e1p"></td>
76
+ <td class="tg-km2t">(1) Building</td>
77
+ <td class="tg-8jgo">8.14</td>
78
+ <td class="tg-8jgo">3.26</td>
79
+ <td class="tg-j3z6"></td>
80
+ <td class="tg-km2t">(6) Coniferous</td>
81
+ <td class="tg-8jgo">2.74</td>
82
+ <td class="tg-8jgo">10.24</td>
83
+ <td class="tg-pop6"></td>
84
+ <td class="tg-km2t">(11) Agricultural Land</td>
85
+ <td class="tg-8jgo">10.98</td>
86
+ <td class="tg-8jgo">18.19</td>
87
+ <td class="tg-7f0h"></td>
88
+ <td class="tg-km2t">(16) Mixed</td>
89
+ <td class="tg-8jgo">0.05</td>
90
+ <td class="tg-8jgo">0.12</td>
91
+ </tr>
92
+ <tr>
93
+ <td class="tg-9efv"></td>
94
+ <td class="tg-km2t">(2) Pervious surface</td>
95
+ <td class="tg-8jgo">8.25</td>
96
+ <td class="tg-8jgo">3.82</td>
97
+ <td class="tg-edjf"></td>
98
+ <td class="tg-km2t">(7) Deciduous</td>
99
+ <td class="tg-8jgo">15.38</td>
100
+ <td class="tg-8jgo">24.79</td>
101
+ <td class="tg-3chm"></td>
102
+ <td class="tg-km2t">(12) Plowed land</td>
103
+ <td class="tg-8jgo">3.88</td>
104
+ <td class="tg-8jgo">1.81</td>
105
+ <td class="tg-40e0"></td>
106
+ <td class="tg-km2t">(17) Ligneous</td>
107
+ <td class="tg-8jgo">0.01</td>
108
+ <td class="tg-8jgo">-</td>
109
+ </tr>
110
+ <tr>
111
+ <td class="tg-3m6m"></td>
112
+ <td class="tg-km2t">(3) Impervious surface</td>
113
+ <td class="tg-8jgo">13.72</td>
114
+ <td class="tg-8jgo">5.87</td>
115
+ <td class="tg-jmwx"></td>
116
+ <td class="tg-km2t">(8) Brushwood</td>
117
+ <td class="tg-8jgo">6.95</td>
118
+ <td class="tg-8jgo">3.81</td>
119
+ <td class="tg-kors"></td>
120
+ <td class="tg-km2t">(13) Swimming pool</td>
121
+ <td class="tg-8jgo">0.01</td>
122
+ <td class="tg-8jgo">0.02</td>
123
+ <td class="tg-qwc7"></td>
124
+ <td class="tg-km2t">(18) Greenhouse</td>
125
+ <td class="tg-8jgo">0.12</td>
126
+ <td class="tg-8jgo">0.15</td>
127
+ </tr>
128
+ <tr>
129
+ <td class="tg-r3rw"></td>
130
+ <td class="tg-km2t">(4) Bare soil</td>
131
+ <td class="tg-8jgo">3.47</td>
132
+ <td class="tg-8jgo">1.6</td>
133
+ <td class="tg-69kt"></td>
134
+ <td class="tg-km2t">(9) Vineyard</td>
135
+ <td class="tg-8jgo">3.13</td>
136
+ <td class="tg-8jgo">2.55</td>
137
+ <td class="tg-oe15"></td>
138
+ <td class="tg-km2t">(14) Snow</td>
139
+ <td class="tg-8jgo">0.15</td>
140
+ <td class="tg-8jgo">-</td>
141
+ <td class="tg-oedl"></td>
142
+ <td class="tg-km2t">(19) Other</td>
143
+ <td class="tg-8jgo">0.14</td>
144
+ <td class="tg-8jgo">0.04</td>
145
+ </tr>
146
+ <tr>
147
+ <td class="tg-9xgv"></td>
148
+ <td class="tg-km2t">(5) Water</td>
149
+ <td class="tg-8jgo">4.88</td>
150
+ <td class="tg-8jgo">3.17</td>
151
+ <td class="tg-0u95"></td>
152
+ <td class="tg-km2t">(10) Herbaceous vegetation</td>
153
+ <td class="tg-8jgo">17.84</td>
154
+ <td class="tg-8jgo">19.76</td>
155
+ <td class="tg-x5zi"></td>
156
+ <td class="tg-km2t">(15) Clear cut</td>
157
+ <td class="tg-8jgo">0.15</td>
158
+ <td class="tg-8jgo">0.82</td>
159
+ <td class="tg-zv4m"></td>
160
+ <td class="tg-zv4m"></td>
161
+ <td class="tg-8jgo"></td>
162
+ <td class="tg-8jgo"></td>
163
+ </tr>
164
  </tbody>
165
  </table>
 
 
 
166
  <br><br>
167
 
168
+ ## Dataset Structure
169
 
170
+ ### Spatio-Temporal Distribution
171
+ The FLAIR dataset consists of 77 762 patches. Each patch includes a high-resolution aerial image of $0.2$\:m (512x512), a yearly satellite image time series with a spatial resolution of 10m (40x40), and pixel-precise elevation and land cover annotations at 0.2m resolution (512x512).
172
 
173
+ ### Annotations
174
+ Each pixel has been manually annotated by photo-interpretation of the 20cm resolution aerial imagery, carried out by a team supervised by geography experts from the IGN.
175
+ Movable objects like cars or boats are annotated according to their underlying cover.
176
 
177
+ ### Training Splits
178
+ The dataset is made up of 50 distinct spatial domains, aligned with the administrative boundaries of the French départements.
179
+ For our experiments, we designate 32 domains for training, 8 for validation, and reserve 10 as the official test set.
180
+ This arrangement ensures a balanced distribution of semantic classes, radiometric attributes, bioclimatic conditions, and acquisition times across each set.
181
+ Consequently, every split accurately reflects the landscape diversity inherent to metropolitan France.
182
+ It is important to mention that the patches come with meta-data permitting alternative splitting schemes, for example focused on domain shifts.
 
 
 
 
 
183
 
184
 
185
  ## Reference
186
+ Please include a citation to the following article if you use the FLAIR dataset:
187
 
188
  ```
189
+ @misc{garioud2023flair,
190
+ title={FLAIR: a Country-Scale Land Cover Semantic Segmentation Dataset From Multi-Source Optical Imagery},
191
+ author={Anatol Garioud and Nicolas Gonthier and Loic Landrieu and Apolline De Wit and Marion Valette and Marc Poupée and Sébastien Giordano and Boris Wattrelos},
192
+ year={2023},
193
+ eprint={2310.13336},
194
+ archivePrefix={arXiv},
195
+ primaryClass={cs.CV}
196
  }
197
  ```
198