What would it be like to take a picture of Lin Huiyin with iPhone? This “through camera” replaces old photos with modern ones

How to take a big picture of Lincoln with iPhone? In addition to crossing the body back to the 19th century, you can also try this “crossing camera”. < / P > < p > since the invention of photography technology for more than 100 years, numerous historical celebrities have left black and white photos. However, due to different times, different technologies and the erosion of time, these historical photos are gradually distorted. < / P > < p > now, the latest paper time travel photography from the University of Washington, UC Berkeley and Google shows the historical figures in these photos on the screen. < / P > < p > it’s nothing new to recover faces from black-and-white photos and even portraits. For example, recently, videos of the restoration of terracotta warriors and horses and ancient emperors have become popular on social networks. < / P > < p > let’s take a picture of Kafka, a famous writer. The photos recovered by Zhang et al. Seem to have a lot of flaws on the skin, while the photos recovered by “through camera” seem to have their own “skin grinding”. < / P > < p > this is not the case. The defect on Kafka’s face was caused by the negative technology of that year. If we just color black and white photos, we can not fully characterize the characteristics and aging process of old films. In 1907, panchromatic negative sensitive to red, blue and green was born. Therefore, when pankster was photographed with new technology in 1913, his facial skin was much cleaner and smoother: < / P > < p > therefore, to restore these historical photos, we should not only consider the problems of blurring, fading, noise, low resolution, but also consider the color sensitivity differences of different old photos. < / P > < p > for the photos taken before 1873, the author used the blue light sensitive model. For the photos taken from 1873 to 1907, the author manually selected between the blue light sensitive and the positive negative, and then for the photos taken after 1907, the author manually selected from all the models. < / P > < p > the author uses the stylegan2 framework to map old photos into the space of modern high-resolution photos, so as to achieve all these effects in a unified framework. < / P > < p > the author uses other color sample images as reference, which have similar facial features to black-and-white images, but contain high-frequency details as well as natural color and lighting. In order to further reduce the perceptual identity gap between the input image and the modern portrait, they also designed a reconstruction method that is especially suitable for old photos. < / P > < p > in order to further limit the color and skin details and make them match the same level images, the model introduces the loss of color conversion lcolor to force the similarity between the output of torgb layer of stylegan2 and the same level images. < / P > < p > in order to further reduce the perceived identity gap between the input image and the modern portrait, the author specially designed a reconstruction loss lrecon suitable for the old photos, which is robust to the defects of the old photos. < / P > < p > according to the author, it takes 10 minutes to run on NVIDIA Titan x to generate such a 1024 × 1024 photo. < / P > < p > in addition, the generation of photos also depends on the selection of statistical images. If you choose a race or gender different from the original photos, black people will become white and women will become men. < / P > < p > the author said in the project home page that the code of “through camera” will soon be open source. However, according to the rhythm of the paper “flooded canteen”, we may have to wait two or three months. Developed a “plug and play” solar power generation scheme, and “5B” won a $12 million round a financing