Claim · model-written summary
Plant detection model performance: this effect is limited and only holds when the new images cover new variation.
Not reportedMaterialGeneCompoundOrgan / stageInteractorInteractor roleSecond interactorSecond interactor roleTreatmentComparatorSettingDirectionValueQualifier
Verbatim
e0“this effect is limited and only holds when the new images cover new variation.”
t0“plant”
Title
Improved generalization of a plant-detection model for precision weed control
Thijs Ruigrok; Eldert J. van Henten; Gert Kootstra Computers and Electronics in Agriculture 2022. doi:10.1016/j.compag.2022.107554
Other claims from this paper (6)
- Method / resource Plant detection model performance: the average generalization error of our plant-detection model was 0.06 mAP
- Method / resource Plant detection model performance: Increasing the number of sub-datasets for training, while keeping the total number of training images constant, increased the variation covered by the training set and improved the generalization of our plant-detection model.
- Method / resource Plant detection model performance: Adding more training images sampled from the same datasets increased the generalization further.
- Method / resource Plant detection model performance: Naively adding more images does not prepare the model for specific scenarios outside the training distribution.
- Method / resource Plant detection model performance: Using incremental training the model can be adapted to such scenarios and the generalization error can be mitigated.
- Method / resource Plant detection model performance: finetuning on as little as 25 images can already mitigate the generalization error.