CharlesGaydon commited on
Commit
efddf95
1 Parent(s): b0804fb

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +13 -5
README.md CHANGED
@@ -30,8 +30,7 @@ This dataset includes 135,569 patches, each measuring 50m*50m, covering a cumula
30
  Each patch represents a monospecific forest, labeled with a single tree species to facilitate classification tasks.
31
  The proposed classification features 13 semantic classes, hierarchically grouping 18 tree species from 9 different tree genus.
32
 
33
- A reference train/val/test split is provided with class stratification.
34
- To account for spatial autocorrelation, each forest exclusively belongs to either the train, validation, or test set.
35
 
36
  | Class | Train (%) | Val (%) | Test (%) |
37
  |-------|------------:|----------:|-----------:|
@@ -64,15 +63,24 @@ Lidar points clouds | Aerial imagery
64
  ## Annotation
65
 
66
  Annotation were made at the forest level, and considering only monospecific forests. A semi-automatic approach was adopted in which pure forest polygons
67
- were selected and then curated by photointerpreters. The annotation polygons came from the [BD Forêt](https://inventaire-forestier.ign.fr/spip.php?article646),
68
  a forest vector database of tree species occupation in France. Ground truths from the F[rench National Forest Inventory](https://inventaire-forestier.ign.fr/?lang=en)
69
- were also used.
70
 
71
  ## Data Splits
72
 
73
  The polygons were sampled in southern France due to the partial availability of the Lidar data at the time of dataset creation.
74
- They are located in 40 distinct French administrative departments.
 
 
 
 
 
 
75
 
 
 
 
76
 
77
  ## Citation
78
  Please include a citation to the following article if you use the PureForest dataset:
 
30
  Each patch represents a monospecific forest, labeled with a single tree species to facilitate classification tasks.
31
  The proposed classification features 13 semantic classes, hierarchically grouping 18 tree species from 9 different tree genus.
32
 
33
+ A reference train/val/test split is provided.
 
34
 
35
  | Class | Train (%) | Val (%) | Test (%) |
36
  |-------|------------:|----------:|-----------:|
 
63
  ## Annotation
64
 
65
  Annotation were made at the forest level, and considering only monospecific forests. A semi-automatic approach was adopted in which pure forest polygons
66
+ were selected and then curated by expert photointerpreters from the IGN. The annotation polygons came from the [BD Forêt](https://inventaire-forestier.ign.fr/spip.php?article646),
67
  a forest vector database of tree species occupation in France. Ground truths from the F[rench National Forest Inventory](https://inventaire-forestier.ign.fr/?lang=en)
68
+ were also used to improve the condidence in the purity of the forests.
69
 
70
  ## Data Splits
71
 
72
  The polygons were sampled in southern France due to the partial availability of the Lidar data at the time of dataset creation.
73
+ They are located in 40 distinct French administrative departments, covering a large diversity of territories and forests.
74
+ To define a common benchmark, we divided the data into train, validation, and test sets, with a stratification on semantic labels.
75
+ Annotation polygons are scattered across southern France, leading to a good geographical diversity within each semantic class.
76
+ To account for the high spatial autocorrelation, the split is performed at the annotation polygon level:
77
+ each forest exclusively belongs to either the train, validation, or test set.
78
+ This makes PureForest suitable to evaluate the territorial generalization of classification models.
79
+ We aimed for a 70%-15%-15% split across the train, validation, and test sets, respectively.
80
 
81
+ Approximate positions of forests in PureForest
82
+
83
+ ![](./dataset_extent_map.excalidraw.png)
84
 
85
  ## Citation
86
  Please include a citation to the following article if you use the PureForest dataset: