Adapte ArcGIS deep learing "Solar Photovoltaic Park Classification - Global" for RBG images (3bands)

Question

Hello everyone,

@RamiAlouta @KateHess

I am currently working on the ArcGIS deep learning model titled "Solar Photovoltaic Park Classification - Global", and I am encountering an issue that I can't seem to resolve.

My images are in RGB format with 3 bands, but the model appears to have been trained on multispectral images with 12 bands. This results in size mismatch errors when I try to load the model weights.

Here is the exact error message I am receiving:

size mismatch for layers.0.0.weight: copying a param with shape torch.Size([64, 12, 7, 7]) from checkpoint, the shape in current model is torch.Size([64, 3, 7, 7]).

Could anyone assist me in adjusting the model to work correctly with RGB images with 3 bands? Any suggestions or experiences with this type of issue would be greatly appreciated.

Thank you very much for your help!

Best regards,

SaranyaM · Answer

Hi,
This model is trained only to work on Sentinel2.

mazzouzi · Answer

Hello @S6

Thank you for your reply. My aim is to use the ESRI model on orthophoto images in RBG 3 bands, not the sentinel images.

I don't know if this is possible

SaranyaM · Answer

Hi,

Can you confirm if you have used the Sentinel- 2  L2A imagery - BOA Reflectance as shown in the below image? In case you have used your own composite, try again with BOA Reflectance - the model should work.