Fig. 2

A Results from the data demand experiment. Average accuracy with 95% confidence intervals is given for both body and wing classification. Performance on the testing set for both transfer learning and fine-tuning is shown. Results from the device comparison experiment are presented as average accuracy with 95% CI. B Performance of models trained on phone, macro and micro, grouped by device. C Performance of body classification models trained on singular devices, categorized by image capture device in the testing set. D Performance of wing classification models trained on singular devices, categorized by image capture device in the testing set