Figure 2
From: Neural network control of focal position during time-lapse microscopy of cells
The effect of downsampling on CNN accuracy. (a–c) Images downsampled by (a) 10X, (b) 4X and (c) 2X from an original image of 1000 × 1000 pixels in size. (d) Error of CNNs trained with different downsampling and batch sizes. The error is defined as the difference between the inferred and annotated z-position. Symbols give the mean error and bars are ±1 standard deviation. Blue dots, orange triangles, and green stars correspond to batch sizes of 10, 25 (22 for 2X downsampling) and 50, respectively. (e) Mean squared error (MSE) with symbols and colors as described in (d). (f) Time taken to train each CNN.
