Table 3. Presented methods in section 5.3.

Name	Reference	Network type	Input	Output
—	Hershko, Weiss, Shechtman, et al. [95]	1) Convolutional (for color channel identification of open-aperture PSFs)	Images of individual open-aperture PSFs	Color channel
		2) Small architecture	Simulated point sources, mismatch between net 3 prediction and ground truth	PSFs, used to optimize SLM pattern in combination with output from net 3
		3) Encoder-decoder (for analysis of PSFs after SLM optimization)	Camera frame with multiple PSFs	Color channel, 2D coordinates of PSFs
DeepSTORM3D	Nehme, Shechtman, et al. [96]	1) Convolutional with skipped connections	Simulated point sources, mismatch between net 2 prediction and ground truth	PSFs, used to optimize phase mask pattern in combination with output from net 2
		2) Convolutional with skipped connections and upsampling	Camera frame with multiple 3D PSFs	3D coordinates of PSFs