When I used the detect object using deep learning using self created training sample. I got the following result that is 100 percent incorrect.
Hi @MārisNitcis
Is there a reason you created your own training samples instead of using the pre-existing Palm Detection model? It is not perfect, but can be expected to be up to 75% accurate when used with the correct imagery (5-15cm resolution RGB imagery).
@MārisNitcis the quality of the model depends on various factors. Two of the most important factors are the training data (should have enough, accurate and diverse) and the hyperparameters (more specifically, number of epochs, and validation data).. The model should be trained with as many possible epochs. But, if the training samples are not enough, training the model for more epochs will not help. You can look the model quality by reviewing the model_metrics.html report. You can use ready to use model as suggested above. Hope this helps.
A side note: We have an ArcGIS Image Analyst dedicated community and here you might get response sooner.
Cheers!
Pavan Yadav | Product Engineer - Imagery and AI
Esri | 380 New York | Redlands, 92373 | USA
https://www.linkedin.com/in/pavan-yadav-1846606/