Inquiry about DeepForest model application on Caribbean drone orthophoto

[JulySuarez]

My name is July Suarez, I am from Colombia and I am currently pursuing a Ph.D. in Marine Sciences.

I am exploring the use of the DeepForest model applied to a drone-derived orthophoto of a vegetation patch in the Colombian Caribbean region. This area presents an interesting structural diversity in terms of tree canopies. However, the canopy segmentation results obtained so far seem unusual and do not appear to make logical sense.

I am attaching a few reference images for your consideration.

I would greatly appreciate any guidance you could provide, based on your experience, regarding potential causes or parameters that might be affecting the model's accuracy — or any specific recommendations for working with this type of imagery or ecosystem.

Thank you in advance for your attention and support.

[bw4sz]

Can you show the code you used to create this set, and if possible, attach a piece of the image. This doesn't feel like a problem with the model, but how the data was processed.

[JulySuarez]

Kind regards.

Thank you very much for your response. I am hereby sharing the download link for the code and a fragment of the image. I have also sent the information to your email in case you experience any issues with the link. Please do not hesitate to contact me should you require any further information.
Link: https://we.tl/t-fEWhAyPsQd

Thank you for your attention.