It’s been few quarters, since then Xiaomi has better its gadgets sensor modules a lot. Their multiple gadgets released this year have garnered good ratings in DxOmark. The latest device to acquire this kind of rating is Xiaomi Mi Mix 3 what carries 12 megapixels +12 megapixel primary scanner on the rear facet even though a 24 megapixel + 2-megapixel sensors sit on the front side. Currently, Xiaomi has posted its fresh algorithm DeepExposure AI which helps to restore poorly exposed Pictures.
Other of the mobile cameras have the problem of poorly exposing shots if the light is brighter or it is too dark. Hardly, except some manufacturers, no one has obtained any success of getting rid of the poorly exposed Pictures. Today, Xiaomi has unveiled DeepExposure AI algorithm what will help balance your images with accurate lights and details.
Xioami has employed AI to fix this issue to higher than or underexposed images. Xiaomi’s DeepExposure balance the exposure by dividing the photo into subfragments and then the exposure price of each fragment is adjusted separately. At some point, different parts are merged to form the final image. You can have a Check at the general process in the image Here.
“The correct exposure is the primary of capturing high-quality photos in computational photography, especially for mobile smartphones that are limited by sizes of digital camera modules,” the researchers wrote. “Inspired by luminosity masks usually applied by professional photographers, in this paper, we develop a novel algorithm for learning local exposures with deep reinforcement adversarial learning.”
The fresh algorithm divides the whole process into image segmentation and “action-generation” stage. Whenever segmentation, the second part starts with input low-resolution, sub-images, and direct fusion of the picture takes place. All the pics are processed with each’s local and global exposures.
The new algorithm is developed in Google’s TensorFlow framework on a Nvidia P40 Tesla GPU. Xiaomi team used the photos from the MIT-Adobe FiveK dataset.