CharlesGaydon
commited on
Commit
•
670d769
1
Parent(s):
dbbef9e
Update README.md
Browse files
README.md
CHANGED
@@ -27,17 +27,17 @@ The hereby PureForest dataset is derived from 449 different forests located in 4
|
|
27 |
It is characterized by two modalities: high density aerial Lidar point clouds with a density of 10 pulses per square meter,
|
28 |
and high resolution aerial imagery with a spatial resolution of 0.2 m.
|
29 |
|
30 |
-
This dataset includes 135,569 patches, each measuring
|
31 |
-
Each patch represents a monospecific forest,
|
32 |
-
The proposed classification
|
33 |
|
34 |
A reference train/val/test split is provided.
|
35 |
|
36 |
## Dataset Structure
|
37 |
<hr style='margin-top:-1em; margin-bottom:0' />
|
38 |
-
The PureForest dataset consists of a total of 135,569 patches: 69111 in the train set, 13523 in the
|
39 |
-
Each patch includes a high-resolution aerial image (
|
40 |
-
Band order is
|
41 |
|
42 |
Lidar and imagery data were acquired over several years in distinct programs, and consequently they are asynchrone: depending on the location, up to 3 years might separate them.
|
43 |
|
@@ -47,9 +47,9 @@ Lidar points clouds | Aerial imagery
|
|
47 |
|
48 |
## Annotations
|
49 |
<hr style='margin-top:-1em; margin-bottom:0' />
|
50 |
-
|
51 |
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),
|
52 |
-
a forest vector database of tree species occupation in France. Ground truths from the
|
53 |
were also used to improve the condidence in the purity of the forests.
|
54 |
|
55 |
| Class | Train (%) | Val (%) | Test (%) |
|
@@ -72,14 +72,13 @@ were also used to improve the condidence in the purity of the forests.
|
|
72 |
<hr style='margin-top:-1em; margin-bottom:0' />
|
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, covering a large diversity of territories and forests.
|
75 |
-
To define a common benchmark, we divided the data into train,
|
76 |
Annotation polygons are scattered across southern France, leading to a good geographical diversity within each semantic class.
|
77 |
-
To account for the high spatial autocorrelation, the split is performed at the annotation polygon level:
|
78 |
-
each forest exclusively belongs to either the train,
|
79 |
This makes PureForest suitable to evaluate the territorial generalization of classification models.
|
80 |
-
We aimed for a 70%-15%-15% split across the train, validation, and test sets, respectively.
|
81 |
|
82 |
-
Approximate positions of forests in PureForest
|
83 |
|
84 |
![](./dataset_extent_map.excalidraw.png)
|
85 |
|
@@ -89,8 +88,8 @@ Please include a citation to the following article if you use the PureForest dat
|
|
89 |
|
90 |
```
|
91 |
@article{gaydon2024pureforest,
|
92 |
-
title={PureForest: A
|
93 |
-
author={Charles Gaydon and
|
94 |
year={2024},
|
95 |
doi={TBD},
|
96 |
}
|
|
|
27 |
It is characterized by two modalities: high density aerial Lidar point clouds with a density of 10 pulses per square meter,
|
28 |
and high resolution aerial imagery with a spatial resolution of 0.2 m.
|
29 |
|
30 |
+
This dataset includes 135,569 patches, each measuring 50 m x 50 m, covering a cumulative exploitable area of 339 km².
|
31 |
+
Each patch represents a monospecific forest, annotated with a single tree species labeled to facilitate classification tasks.
|
32 |
+
The proposed classification has 13 semantic classes, hierarchically grouping 18 tree species from 9 different tree genera.
|
33 |
|
34 |
A reference train/val/test split is provided.
|
35 |
|
36 |
## Dataset Structure
|
37 |
<hr style='margin-top:-1em; margin-bottom:0' />
|
38 |
+
The PureForest dataset consists of a total of 135,569 patches: 69111 in the train set, 13523 in the val set, and 52935 in the test set.
|
39 |
+
Each patch includes a high-resolution aerial image (250 pixels x 250 pixels) at 0.2 m resolution, and a point cloud of high density aerial Lidar (10 pulses/m², ~40pts/m²).
|
40 |
+
Band order is near-infrared, red, greeb, blue. For convenience, the Lidar point clouds are vertically colorized with the aerial images.
|
41 |
|
42 |
Lidar and imagery data were acquired over several years in distinct programs, and consequently they are asynchrone: depending on the location, up to 3 years might separate them.
|
43 |
|
|
|
47 |
|
48 |
## Annotations
|
49 |
<hr style='margin-top:-1em; margin-bottom:0' />
|
50 |
+
Annotations were made at the forest level, and considering only monospecific forests. A semi-automatic approach was adopted in which pure forest polygons
|
51 |
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),
|
52 |
+
a forest vector database of tree species occupation in France. Ground truths from the [French National Forest Inventory](https://inventaire-forestier.ign.fr/?lang=en)
|
53 |
were also used to improve the condidence in the purity of the forests.
|
54 |
|
55 |
| Class | Train (%) | Val (%) | Test (%) |
|
|
|
72 |
<hr style='margin-top:-1em; margin-bottom:0' />
|
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, covering a large diversity of territories and forests.
|
75 |
+
To define a common benchmark, we divided the data into train, val, and test sets, with a stratification on semantic labels.
|
76 |
Annotation polygons are scattered across southern France, leading to a good geographical diversity within each semantic class.
|
77 |
+
To account for the high spatial autocorrelation, the 70%-15%-15% split is performed at the annotation polygon level:
|
78 |
+
each forest exclusively belongs to either the train, val, or test set.
|
79 |
This makes PureForest suitable to evaluate the territorial generalization of classification models.
|
|
|
80 |
|
81 |
+
Approximate positions of forests in PureForest:
|
82 |
|
83 |
![](./dataset_extent_map.excalidraw.png)
|
84 |
|
|
|
88 |
|
89 |
```
|
90 |
@article{gaydon2024pureforest,
|
91 |
+
title={PureForest: A Novel Dataset and Benchmarks for Tree Species Classification in Monospecific Forests from Aerial Lidar and Aerial Images},
|
92 |
+
author={Charles Gaydon and Floryne Roche},
|
93 |
year={2024},
|
94 |
doi={TBD},
|
95 |
}
|