Table 1.

Segmentation network, where b is the batch size and h, w is the size of image patch extracted from AOSLO images.

Layer	Input size	Output size
3 × 3 convolution	(b, h, w, 1)	(b, h, w, 1)
32 unit MDLSTM	(b, h, w, 1)	(b, h, w, 4*32)
3 × 3 convolution	(b, h, w, 4*32)	(b, h, w, 1)
32 unit MDLSTM	(b, h, w, 1)	(b, h, w, 4*32)
Fully connected - hidden	(b, h, w, 4*32)	(b, h, w, 64)
Fully connected - output	(b, h, w, 64)	(b, h, w, 2)
Softmax	(b, h, w, 2)	(b, h, w, 2)