Datasets:

Formats:
csv
Languages:
English
Size:
< 1K
ArXiv:
Libraries:
Datasets
pandas
License:
arjunrao2000 commited on
Commit
0795903
Β·
verified Β·
1 Parent(s): 3c8b460

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +66 -6
README.md CHANGED
@@ -54,12 +54,72 @@ configs:
54
  This dataset card contains usage instructions and metadata for all data-products released with our paper:
55
  *Using Multiple Input Modalities can Improve Data-Efficiency and O.O.D. Generalization for ML with Satellite Imagery.* We release 3 modified versions of 3 benchmark datasets spanning land-cover segmentation, tree-cover regression, and multi-label land-cover classification tasks. These datasets are augmented with auxiliary, geographic inputs. A full list of contributed data products is shown in the table below.
56
 
57
- | **Dataset** | **Task Description** | **Multispectral Input** | **Model** | **Additional Data Layers** | **OOD Test Set Present?** |
58
- |--------------------------------------|------------------------------------|-----------------------------|------------|-------------------------------------------------------|---------------------------|
59
- | [SustainBench](https://arxiv.org/abs/2111.04724) | Farmland boundary delineation | Sentinel-2 RGB | U-Net | OSM rasters, EU-DEM | βœ— |
60
- | [EnviroAtlas](https://arxiv.org/abs/2202.14000) | Land-cover segmentation | NAIP RGB + NIR | FCN | [Prior](https://arxiv.org/abs/2202.14000), OSM rasters | βœ“ |
61
- | [BigEarthNet v2.0](https://bigearth.net/static/documents/Description_BigEarthNet_v2.pdf) | Land-cover classification | Sentinel-2 (10 bands) | ViT | [SatCLIP](https://arxiv.org/abs/2311.17179) embeddings | βœ“ |
62
- | [USAVars](https://arxiv.org/abs/2010.08168) | Tree-cover regression | NAIP RGB + NIR | ResNet-50 | OSM rasters | βœ— |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
63
 
64
  ## Usage Instructions
65
  * Download the `.h5.gz` files in `data/<source dataset name>`. Our source datasets include SustainBench, USAVars, and BigEarthNet2.0. Each dataset with the augmented geographic inputs is detailed in [this section πŸ“¦](#geolayersused)
 
54
  This dataset card contains usage instructions and metadata for all data-products released with our paper:
55
  *Using Multiple Input Modalities can Improve Data-Efficiency and O.O.D. Generalization for ML with Satellite Imagery.* We release 3 modified versions of 3 benchmark datasets spanning land-cover segmentation, tree-cover regression, and multi-label land-cover classification tasks. These datasets are augmented with auxiliary, geographic inputs. A full list of contributed data products is shown in the table below.
56
 
57
+ <table>
58
+ <thead>
59
+ <tr>
60
+ <th>Dataset</th>
61
+ <th>Task Description</th>
62
+ <th>Multispectral Input</th>
63
+ <th>Model</th>
64
+ <th>Additional Data Layers</th>
65
+ <th colspan="2">Dataset Size</th>
66
+ <th>OOD Test Set Present?</th>
67
+ </tr>
68
+ <tr>
69
+ <th></th>
70
+ <th></th>
71
+ <th></th>
72
+ <th></th>
73
+ <th></th>
74
+ <th>Compressed</th>
75
+ <th>Uncompressed</th>
76
+ <th></th>
77
+ </tr>
78
+ </thead>
79
+ <tbody>
80
+ <tr>
81
+ <td><a href="https://arxiv.org/abs/2111.04724">SustainBench</a></td>
82
+ <td>Farmland boundary delineation</td>
83
+ <td>Sentinel-2 RGB</td>
84
+ <td>U-Net</td>
85
+ <td>OSM rasters, EU-DEM</td>
86
+ <td>1.76 GB</td>
87
+ <td>1.78 GB</td>
88
+ <td>βœ—</td>
89
+ </tr>
90
+ <tr>
91
+ <td><a href="https://arxiv.org/abs/2202.14000">EnviroAtlas</a></td>
92
+ <td>Land-cover segmentation</td>
93
+ <td>NAIP RGB + NIR</td>
94
+ <td>FCN</td>
95
+ <td><a href="https://arxiv.org/abs/2202.14000">Prior</a>, OSM rasters</td>
96
+ <td>N/A</td>
97
+ <td>N/A</td>
98
+ <td>βœ“</td>
99
+ </tr>
100
+ <tr>
101
+ <td><a href="https://bigearth.net/static/documents/Description_BigEarthNet_v2.pdf">BigEarthNet v2.0</a></td>
102
+ <td>Land-cover classification</td>
103
+ <td>Sentinel-2 (10 bands)</td>
104
+ <td>ViT</td>
105
+ <td><a href="https://arxiv.org/abs/2311.17179">SatCLIP</a> embeddings</td>
106
+ <td>120 GB (raw), 91 GB (H5)</td>
107
+ <td>205 GB (raw), 259 GB (H5) </td>
108
+ <td>βœ“</td>
109
+ </tr>
110
+ <tr>
111
+ <td><a href="https://arxiv.org/abs/2010.08168">USAVars</a></td>
112
+ <td>Tree-cover regression</td>
113
+ <td>NAIP RGB + NIR</td>
114
+ <td>ResNet-50</td>
115
+ <td>OSM rasters</td>
116
+ <td> 23.56 GB </td>
117
+ <td> 167 GB</td>
118
+ <td>βœ—</td>
119
+ </tr>
120
+ </tbody>
121
+ </table>
122
+
123
 
124
  ## Usage Instructions
125
  * Download the `.h5.gz` files in `data/<source dataset name>`. Our source datasets include SustainBench, USAVars, and BigEarthNet2.0. Each dataset with the augmented geographic inputs is detailed in [this section πŸ“¦](#geolayersused)