Deep learning, pre-trained Sentinel Model for fine-tuning

[santhosh]

Hi , i am trying to train data model using Land Cover Classification (Sentinel-2) model for my study area, i followed the steps in fine tune page .

https://doc.arcgis.com/en/pretrained-models/latest/imagery/finetuning-the-land-cover-classification-...

And for Input Feature Class Or Classified Raster Or Table: I used polygon data 

But i am getting this error when i try to train the model with trained data

Fine tune error using polygon

:Error(s) in loading state_dict for DynamicUnet:

size mismatch for layers.11.0.weight: copying a param with shape torch.Size([17, 108, 1, 1]) from checkpoint, the shape in current model is torch.Size([2, 108, 1, 1]).

size mismatch for layers.11.0.bias: copying a param with shape torch.Size([17]) from checkpoint, the shape in current model is torch.Size([2])..Training was not successful.

[ShivaniPathak]

Hi Santhosh,

Land Cover Classification (Sentinel-2) pretrained model is trained with 16 lulc classes but from the error it seems you are trying to train the model with data having only 2 classes. For finetuning any pretrained model it is expected to train the model on same number of classes. 

These are the classes on which the model is trained:

The model is trained on Corine LULC data