Fig. 1. Network architecture. The encoder consists of a ViT and an adapter for accelerated training. Intensity, depth, and width are predicted by three decoders to finally reconstruct the raw data.

Fig. 2. Network modules design. (a) Convert the microscopic imaging stack into patches by a 3D convolution; (b) prediction head design for intensity, depth, and width.

Fig. 3. Examples of raw measurement data. Scale bar is 50 µm.

Fig. 4. Comparative Experiment 1. (a) Confocal microscope intensity image, with the yellow area indicating the field of view for subsequent analysis; (b) depth image obtained from 1× mode data using the traditional commercial algorithm; (c) depth image obtained from 1× mode data using SSL Depth; (d) depth image obtained from 4× mode data using the traditional commercial algorithm; (e) depth image obtained from 4× mode data using SSL Depth; (f) depth image obtained from 16× mode data using SSL Depth; (g) mean absolute error (L₁) corresponding to the cross-sectional line, assuming that the traditional commercial 1× result is true. The red dashed line shows the error at 4× speed for the commercial microscope. Scale bar is 50 µm.

Fig. 5. Comparative Experiment 2. (a) Confocal microscope intensity image, with the yellow area indicating the field of view for subsequent analysis; (b) depth image obtained from 1× mode data using the traditional commercial algorithm; (c) depth image obtained from 1× mode data using SSL Depth. Scale bar is 50 µm.