DSLR-Quality Photos on Mobile Devices with Deep Convolutional Networks
read more
Citations
Deep Learning in Mobile and Wireless Networking: A Survey
Underexposed Photo Enhancement Using Deep Illumination Estimation
Deep Photo Enhancer: Unpaired Learning for Image Enhancement from Photographs with GANs
Learning Enriched Features for Real Image Restoration and Enhancement
Fast Underwater Image Enhancement for Improved Visual Perception
References
Adam: A Method for Stochastic Optimization
Distinctive Image Features from Scale-Invariant Keypoints
Image quality assessment: from error visibility to structural similarity
Generative Adversarial Nets
Image-to-Image Translation with Conditional Adversarial Networks
Related Papers (5)
Frequently Asked Questions (15)
Q2. What is the probability of the input image being taken by the target camera?
A sigmoidal activation function is applied to the outputs of the last fullyconnected layer containing 1024 neurons and produces a probability that the input image was taken by the target DSLR camera.
Q3. What are the two metrics used in the study?
The authors use classical distance metrics, namely PSNR and SSIM scores: the former measures signal distortion w.r.t. the reference, the latter measures structural similarity which is known to be a strong cue for perceived quality [22].
Q4. What is the loss function for the image enhancement task?
The authors build their loss function under the assumption that the overall perceptual image quality can be decomposed into three independent parts: i) color quality, ii) texture quality and iii) content quality.
Q5. What is the way to achieve the results on this task?
the best photo-realistic results on this task are achieved using a VGG-based loss function [9] and adversarial networks [12] that turned out to be efficient at recovering plausible high-frequency components.
Q6. What are the advantages of larger sensors and high-aperture optics?
Larger sensors and high-aperture optics yield better photo resolution, color rendition and less noise, whereas their additional sensors help to fine-tune shooting parameters.
Q7. What is the sigmoidal activation function used in the transformation network?
All layers in the transformation network have 64 channels and are followed by a ReLU activation function, except for the last one, where a scaled tanh is applied to the outputs.
Q8. What are the two common artifacts that can appear on the processed images?
Two typical artifacts that can appear on the processed images are color deviations (see ground/mountains in first image of Fig. 12) and too high contrast levels (second image).
Q9. What are the main problems of automatic image enhancement?
While a number of photographer tools for automatic image enhancement exist, they are usually focused on adjusting only global parameters such as contrast or brightness, without improving texture quality or taking image semantics into account.
Q10. What is the way to train and evaluate a photo enhancement method?
To train and evaluate their method the authors introduced DPED – a large-scale dataset that consists of real photos captured from three different phones and one high-end reflex camera, and suggested an efficient way of calibrating the images so that they are suitable for image-to-image learning.
Q11. What is the main approach to photo post-processing?
the dominant approach to photo post-processing is still based on manual image correction using specialized retouching software.
Q12. What is the way to achieve better performance on this task?
Considerably better performance on this task was obtained using generative adversarial networks [8] or a 16-layer CNN with a multinomial cross-entropy loss function [27].
Q13. What did the authors find out about the patch size?
Their preliminary experiments revealed that larger patch sizes do not lead to better performance, while requiring considerably more computational resources.
Q14. What is the way to measure the color difference between the enhanced and target images?
To measure the color difference between the enhanced and target images, the authors propose applying a Gaussian blur (see Figure 5) and computing Euclidean distance between the obtained representations.
Q15. What is the effect of the color loss on the enhanced image?
As a result, color loss forces the enhanced image to have the same color distribution as the target one, while being tolerant to small mismatches.