Fig. 4: Integrated microscope equipped in a cell phone for real-time extended depth-of-field imaging.
From: Large depth-of-field ultra-compact microscope by progressive optimization and deep learning
a Left, assemble the proposed integrated microscope in a cell phone. Right, the zoom-in panel shows the integrated microscope module (top) and 3D schematic diagram of the ring-shaped LED that is used for illumination (bottom). b After the introduction of network pruning, we reduced the network parameters by 78% with similar reconstruction performance regarding structure similarity index (SSIM), peak signal-to-noise ratio (PSNR), and perceptual loss (Learned Perceptual Image Patch Similarity, LPIPS). Central line inside the box: Median. Box: interquartile range. Whiskers: Maximum and minimum. Outliers: Individual data points. n = 19 samples. c Comparisons of the rendering time costs for n = 19 samples before and after network pruning. Data points are overlaid. Height of bars: Mean. Error bars: SD. d Left, comparisons of the captures obtained by the integrated microscope (bottom right) and a conventional microscope (top left) on yellow flowers. Right, zoom-in panels of the white dashed boxes on the left. White arrows indicate the structures which are hard to be resolved in a conventional microscope. Representative data from 53 samples. e The same as (d) but on samples of leaves. Representative data from 53 samples.
