Train model to ignore certain objects without labeling "True" Objects

[frigusgulo]

I am getting false positives in a current model I have developed. While I have data that includes commonly labeled "False Positive" features, I was hoping to avoid labeling all the "True Positive" instances by way of creating a dataset with "ignore" masks segmenting the "False Positive" instances and not labeling the "True Positives"

For context, I have a model that segments trees from the view of a timber harvester. My goal is to have the model ignore the harvesting equipment and just label trees.

The dataset without the harvester boom is far superior and thus do not want to include annotation instances of the trees from the "Harvester boom" dataset