Table 1 Comparison of the required times to generate training data

From: Improving 3D deep learning segmentation with biophysically motivated cell synthesis

Method

Manual annotation

Computation preparation

Computation inference

Maximum

37.5 h

-

-

SimOptiGAN

0.4 h

99.5 h

<0.1 h

SimOptiGAN+

4.4 h

112.5 h

148 h

Mem2NucGAN-P

4 h

33.5 h

148 h

Mem2NucGAN-U

4 h

72.5 h

148 h

  1. The Cellpose nuclei segmentation model is not included, as no training data is required for the usage of this model. Manual annotation defines the time required to manually generate or correct label masks. Preparation and inference time include the computation time required for the generation of synthetic images. While preparation includes one-time calculations such as parameter estimation for biophysical simulation and GAN-based model training, data generation comprises the remaining calculation time to generate four synthetic images.
Back to article page