Replacing Mobile Camera ISP with a Single Deep Learning Model

Abstract

As the popularity of mobile photography is growing constantly, lots of efforts are being invested now into building complex hand-crafted camera ISP solutions. In this work, we demonstrate that even the most sophisticated ISP pipelines can be replaced with a single end-to-end deep learning model trained without any prior knowledge about the sensor and optics used in a particular device. For this, we present PyNET, a novel pyramidal CNN architecture designed for fine-grained image restoration that implicitly learns to perform all ISP steps such as image demosaicing, denoising, white balancing, color and contrast correction, demoireing, etc. The model is trained to convert RAW Bayer data obtained directly from mobile camera sensor into photos captured with a professional high-end DSLR camera, making the solution independent of any particular mobile ISP implementation. To validate the proposed approach on the real data, we collected a large-scale dataset consisting of 10 thousand full-resolution RAW-RGB image pairs captured in the wild with the Huawei P20 cameraphone (12.3 MP Sony Exmor IMX380 sensor) and Canon 5D Mark IV DSLR. The experiments demonstrate that the proposed solution can easily get to the level of the embedded P20's ISP pipeline that, unlike our approach, is combining the data from two (RGB + B/W) camera sensors. The dataset, pretrained models and codes used in this paper are available on the project website: https://people.ee.ethz. ch/˜ihnatova/pynet.html

A Deep Learning Approach to Mobile Camera Image Signal Processing

“…The training and evaluation are performed in image pairs from the same smartphone model. More recently, PyNET [11] was proposed with a pyramidal CNN architecture that learns the ISP steps. It processes the image in different scales combining global and local features.…”

“…There are many possible choices for architectures when building convolutional networks. In [11], the authors proposes a network to process RAW images and output RGB images. However, running that network is very costly and slow.…”

“…Due to requirement of deploying our model on smartphones devices and the image-to-image nature of our problem, we decided to use the UNet [2] architecture. This architecture as lighter when compared to the PyNet [11] and still is able to produce high-quality results. The UNet has a sequence of double convolutional layers in two stages.…”

Image Pre-Processing Method of Machine Learning for Edge Detection with Image Signal Processor Enhancement

“…After the invention of camera, the quality of image from machinery has been continuously improved and it is easy to access the image data. It is recognized as the main data itself and is used to extract additional information through complex data processing using artificial intelligence (AI) [ 1 ].…”

PyNET-CA: Enhanced PyNET with Channel Attention for End-to-End Mobile Image Signal Processing

“…We consider a recently proposed end-to-end RAW to RGB reconstruction model PyNET [7], which is basically a CNN designed to exploit both global and local features of the input data. Despite the fact that the PyNET achieves state-of-the-art performance in RAW to RGB reconstruction, there exists some drawbacks in the model architecture and the training process that can be further improved.…”