The `create_savedmodel_simple_fcn.py` script enables you to create a fully convolutional model which does not use any stride.
...
...
@@ -122,7 +122,7 @@ otbcli_TensorflowModelServe \
-out $output_classif
```
## M3Fusion Model
## M3Fusion Model <a name="part3"></a>
The M3Fusion model (stands for MultiScale/Multimodal/Multitemporal satellite data fusion) is a model designed to input time series and very high resolution images.
...
...
@@ -191,7 +191,7 @@ otbcli_TensorflowModelServe \
-output.names "prediction" -out $output_classif
```
## Maggiori model
## Maggiori model <a name="part4"></a>
This architecture was one of the first to introduce a fully convolutional model suited for large scale remote sensing images.
...
...
@@ -242,7 +242,7 @@ otbcli_TensorflowModelServe \
-out $output_classif
```
## Fully convolutional network with separate Pan/MS channels
## Fully convolutional network with separate Pan/MS channels <a name="part5"></a>
It's common that very high resolution products are composed with a panchromatic channel at high-resolution (Pan), and a multispectral image generally at lower resolution (MS).
This model inputs separately the two sources (Pan and MS) separately.