blumenstiel commited on
Commit
4469ce0
·
1 Parent(s): da261a4

Update README

Browse files
Files changed (2) hide show
  1. README.md +7 -2
  2. assets/radar.png +2 -2
README.md CHANGED
@@ -3,6 +3,7 @@ license: cc-by-sa-4.0
3
  pretty_name: TerraMesh
4
  size_categories:
5
  - 10M<n<100M
 
6
  ---
7
 
8
  # TerraMesh
@@ -10,7 +11,7 @@ size_categories:
10
  > **A planetary‑scale, multimodal analysis‑ready dataset for Earth‑Observation foundation models.**
11
 
12
 
13
- **TerraMesh** merges data from **Sentinel‑1 SAR, Sentinel‑2 optical, Copernicus DEM, NDVI and land‑cover** sources into more than **9 million co‑registered patches** ready for large‑scale representation learning. Compared with earlier public corpora it offers **>20×** more pixel values and truly global, multi‑season coverage.
14
 
15
  **Dataset release – End of June.**
16
 
@@ -48,7 +49,11 @@ Heat map of the sample count in a one-degree grid. | Monthly distribution of al
48
 
49
  ## Performance evaluation
50
 
51
- ![radar.png](assets%2Fradar.png)[radar.pdf](assets%2Fradar.pdf)
 
 
 
 
52
 
53
  ---
54
 
 
3
  pretty_name: TerraMesh
4
  size_categories:
5
  - 10M<n<100M
6
+ viewer: false
7
  ---
8
 
9
  # TerraMesh
 
11
  > **A planetary‑scale, multimodal analysis‑ready dataset for Earth‑Observation foundation models.**
12
 
13
 
14
+ **TerraMesh** merges data from **Sentinel‑1 SAR, Sentinel‑2 optical, Copernicus DEM, NDVI and land‑cover** sources into more than **9 million co‑registered patches** ready for large‑scale representation learning.
15
 
16
  **Dataset release – End of June.**
17
 
 
49
 
50
  ## Performance evaluation
51
 
52
+ ![radar.png](assets%2Fradar.png)
53
+
54
+ TerraMesh was used to pre-train [TerraMind-B](https://huggingface.co/ibm-esa-geospatial/TerraMind-1.0-base).
55
+ On the six evaluated segmentation tasks from PANGAEA bench, TerraMind‑B reaches an average mIoU of 66.6%, the best overall score with an average rank of 2.33. This amounts to roughly a 3pp improvement over the next‑best open model (CROMA), underscoring the benefits of pre‑training on TerraMesh.
56
+ Compared to an ablation model pre-trained only on SSL4EO-S12 locations TerraMind-B performs overall 1pp better with better global generalization on more remote tasks like CTM-SS.
57
 
58
  ---
59
 
assets/radar.png CHANGED

Git LFS Details

  • SHA256: 9e86a435a324fafe848c4e4e5e693038b2f0ebf1e0236e6ffb539c41497bb920
  • Pointer size: 132 Bytes
  • Size of remote file: 1.44 MB

Git LFS Details

  • SHA256: 374dd3594c27c09edbd0af8c88ab320607513b6fc4c7b83ed53fe253fc3624ab
  • Pointer size: 131 Bytes
  • Size of remote file: 576 kB