Update README.md
Browse files
README.md
CHANGED
@@ -114,15 +114,15 @@ Although the model can be applied to other type of very high spatial earth obser
|
|
114 |
The product called ([BD ORTHO®](https://geoservices.ign.fr/bdortho)) has its own spatial and radiometric specifications. The model is not intended to be generic to other type of very high spatial resolution images but specific to BD ORTHO images.
|
115 |
As a result, the prediction produced by the model would be all the better as the user images are similar to the original ones.
|
116 |
|
117 |
-
**Radiometry of input images** :
|
118 |
The BD ORTHO input images are distributed in 8-bit encoding format per channel. When traning the model, input normalization was performed (see section **Traing Details**).
|
119 |
It is recommended that the user apply the same type of input normalization while inferring the model.
|
120 |
|
121 |
-
**Multi-domain model** :
|
122 |
The FLAIR-INC dataset that was used for training is composed of 75 radiometric domains. In the case of aerial images, domain shifts are frequent and are mainly due to : the date of acquisition of the aerial survey (april to november), the spatial domain (equivalent to a french department administrative division) and downstream radimetric processing.
|
123 |
By construction (sampling 75 domains) the model is robust to these shifts, and can be applied to any images of the ([BD ORTHO® product](https://geoservices.ign.fr/bdortho)).
|
124 |
|
125 |
-
**Land Cover classes of prediction** :
|
126 |
The orginial class nomenclature of the FLAIR Dataset is made up of 19 classes(See the [FLAIR dataset](https://huggingface.co/datasets/IGNF/FLAIR) page for details).
|
127 |
However 3 classes corresponding to uncertain labelisation (Mixed (16), Ligneous (17) and Other (19)) and 1 class with very poor labelling (Clear cut (15)) were deasctivated during training.
|
128 |
As a result, the logits produced by the model are of size 19x1, but class 15,16,17 and 19 : (1) should appear at 0 in the logits (2) should never predicted in the Argmax.
|
@@ -133,15 +133,15 @@ As a result, the logits produced by the model are of size 19x1, but class 15,16,
|
|
133 |
|
134 |
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
|
135 |
|
136 |
-
**Using the model on input images with other spatial resolution** :
|
137 |
The FLAIR-INC_RVBIE_resnet34_unet_15cl_norm model was trained with fixed scale conditions.All patches used for training are derived from aerial images of 0.2 meters spatial resolution. Only flip and rotate augmentation were performed during the training process.
|
138 |
No data augmentation method concerning scale change was used during training. The user should pay attention that generalization issues can occur while applying this model to images that have different spatial resolutions.
|
139 |
|
140 |
-
**Using the model for other remote sensing sensors** :
|
141 |
The FLAIR-INC_RVBIE_resnet34_unet_15cl_norm model was trained with aerial images of the ([BD ORTHO® product](https://geoservices.ign.fr/bdortho)) that encopass very specific radiometric image processing.
|
142 |
Using the model on other type of aerial images or satellite images may imply the use of transfer learning or domain adaptation techniques.
|
143 |
|
144 |
-
**Using the model on other spatial areas** :
|
145 |
The FLAIR-INC_RVBIE_resnet34_unet_15cl_norm model was trained on patches reprensenting the French Metropolitan territory.
|
146 |
The user should be aware that applying the model to other type of landscapes may imply a drop in model metrics.
|
147 |
|
@@ -166,7 +166,7 @@ Use the code below to get started with the model.
|
|
166 |
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
|
167 |
218 400 patches of 512 x 512 pixels were used to train the **FLAIR-INC_RVBIE_resnet34_unet_15cl_norm** model.
|
168 |
The train/validation split was performed patchwise to obtain a 80% / 20% distribution between train and validation.
|
169 |
-
Annotation was performed at the _zone_ level (~100 patches per _zone_). Spatial independancy between patches is guaranted as patches from the same _zone_ were assigned to the same
|
170 |
Here are the number of patches used for train and validation :
|
171 |
| TRAIN set | 174 700 patches |
|
172 |
| VALIDATION set | 43 700 patchs |
|
@@ -179,20 +179,19 @@ Here are the number of patches used for train and validation :
|
|
179 |
|
180 |
#### Preprocessing [optional]
|
181 |
|
182 |
-
For traning the model, input normalization was performed so as the input dataset has a mean
|
183 |
-
|
184 |
|
185 |
-
Input normalization was performed
|
186 |
| Modalities | Mean (Train + Validation) |Std (Train + Validation) |
|
187 |
| ----------------------- | ----------- |----------- |
|
188 |
| Red Channel (R) | 105.08 |52.17 |
|
189 |
-
| Green Channel (
|
190 |
| Blue Channel (B) | 101.82 |44.00 |
|
191 |
| Infrared Channel (I) | 106.38 |39.69 |
|
192 |
| Elevation Channel (E) | 53.26 |79.30 |
|
193 |
|
194 |
|
195 |
-
{{ preprocessing | default("[More Information Needed]", true)}}
|
196 |
|
197 |
|
198 |
#### Training Hyperparameters
|
|
|
114 |
The product called ([BD ORTHO®](https://geoservices.ign.fr/bdortho)) has its own spatial and radiometric specifications. The model is not intended to be generic to other type of very high spatial resolution images but specific to BD ORTHO images.
|
115 |
As a result, the prediction produced by the model would be all the better as the user images are similar to the original ones.
|
116 |
|
117 |
+
_**Radiometry of input images**_ :
|
118 |
The BD ORTHO input images are distributed in 8-bit encoding format per channel. When traning the model, input normalization was performed (see section **Traing Details**).
|
119 |
It is recommended that the user apply the same type of input normalization while inferring the model.
|
120 |
|
121 |
+
_**Multi-domain model**_ :
|
122 |
The FLAIR-INC dataset that was used for training is composed of 75 radiometric domains. In the case of aerial images, domain shifts are frequent and are mainly due to : the date of acquisition of the aerial survey (april to november), the spatial domain (equivalent to a french department administrative division) and downstream radimetric processing.
|
123 |
By construction (sampling 75 domains) the model is robust to these shifts, and can be applied to any images of the ([BD ORTHO® product](https://geoservices.ign.fr/bdortho)).
|
124 |
|
125 |
+
_**Land Cover classes of prediction**_ :
|
126 |
The orginial class nomenclature of the FLAIR Dataset is made up of 19 classes(See the [FLAIR dataset](https://huggingface.co/datasets/IGNF/FLAIR) page for details).
|
127 |
However 3 classes corresponding to uncertain labelisation (Mixed (16), Ligneous (17) and Other (19)) and 1 class with very poor labelling (Clear cut (15)) were deasctivated during training.
|
128 |
As a result, the logits produced by the model are of size 19x1, but class 15,16,17 and 19 : (1) should appear at 0 in the logits (2) should never predicted in the Argmax.
|
|
|
133 |
|
134 |
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
|
135 |
|
136 |
+
_**Using the model on input images with other spatial resolution**_ :
|
137 |
The FLAIR-INC_RVBIE_resnet34_unet_15cl_norm model was trained with fixed scale conditions.All patches used for training are derived from aerial images of 0.2 meters spatial resolution. Only flip and rotate augmentation were performed during the training process.
|
138 |
No data augmentation method concerning scale change was used during training. The user should pay attention that generalization issues can occur while applying this model to images that have different spatial resolutions.
|
139 |
|
140 |
+
_**Using the model for other remote sensing sensors**_ :
|
141 |
The FLAIR-INC_RVBIE_resnet34_unet_15cl_norm model was trained with aerial images of the ([BD ORTHO® product](https://geoservices.ign.fr/bdortho)) that encopass very specific radiometric image processing.
|
142 |
Using the model on other type of aerial images or satellite images may imply the use of transfer learning or domain adaptation techniques.
|
143 |
|
144 |
+
_**Using the model on other spatial areas**_ :
|
145 |
The FLAIR-INC_RVBIE_resnet34_unet_15cl_norm model was trained on patches reprensenting the French Metropolitan territory.
|
146 |
The user should be aware that applying the model to other type of landscapes may imply a drop in model metrics.
|
147 |
|
|
|
166 |
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
|
167 |
218 400 patches of 512 x 512 pixels were used to train the **FLAIR-INC_RVBIE_resnet34_unet_15cl_norm** model.
|
168 |
The train/validation split was performed patchwise to obtain a 80% / 20% distribution between train and validation.
|
169 |
+
Annotation was performed at the _zone_ level (~100 patches per _zone_). Spatial independancy between patches is guaranted as patches from the same _zone_ were assigned to the same set (TRAIN or VALIDATION).
|
170 |
Here are the number of patches used for train and validation :
|
171 |
| TRAIN set | 174 700 patches |
|
172 |
| VALIDATION set | 43 700 patchs |
|
|
|
179 |
|
180 |
#### Preprocessing [optional]
|
181 |
|
182 |
+
For traning the model, input normalization was performed so as the input dataset has **a mean=0** and a **standard deviation = 1** channel wise.
|
183 |
+
We used the statistics of TRAIN+VALIDATION for input normalization. It is recommended that the user apply the same type of input normalization. Here are the statistics of the TRAIN+VALIDATIOn set :
|
184 |
|
|
|
185 |
| Modalities | Mean (Train + Validation) |Std (Train + Validation) |
|
186 |
| ----------------------- | ----------- |----------- |
|
187 |
| Red Channel (R) | 105.08 |52.17 |
|
188 |
+
| Green Channel (G) | 110.87 |45.38 |
|
189 |
| Blue Channel (B) | 101.82 |44.00 |
|
190 |
| Infrared Channel (I) | 106.38 |39.69 |
|
191 |
| Elevation Channel (E) | 53.26 |79.30 |
|
192 |
|
193 |
|
194 |
+
<!--{{ preprocessing | default("[More Information Needed]", true)}} -->
|
195 |
|
196 |
|
197 |
#### Training Hyperparameters
|