Table 2 Counting metrics, mean absolute error (MAE) (Eq. 10) and symmetric Mean Absolute Percentage Error (sMAPE) (Eq. 11), for the DM and SNDM model (in MC dropout) in comparison with other counting methods that use popular R-CNN object detectors: Faster37 and Cascade38, which are discussed by Majchrowska et al.23.
From: Self-normalized density map (SNDM) for counting microbiological objects
Model | Detection metrics | |
|---|---|---|
MAE | sMAPE | |
DM | 5.74 | 12.58% |
Self-normalized DM | 3.65 | 4.05% |
Faster R-CNN | 4.75 | 5.32% |
Cascade R-CNN | 4.31 | 4.86 % |
