English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Deep learning (DL) has been applied extensively in many computational imagingproblems, often leading to superior performance over traditional iterativeapproaches. However, two important questions remain largely unanswered: first,how well can the trained neural network generalize to objects very differentfrom the ones in training? This is particularly important in practice, sincelarge-scale annotated examples similar to those of interest are often notavailable during training. Second, has the trained neural network learnt theunderlying (inverse) physics model, or has it merely done something trivial,such as memorizing the examples or point-wise pattern matching? This pertainsto the interpretability of machine-learning based algorithms. In this work, weuse the Phase Extraction Neural Network (PhENN), a deep neural network (DNN)for quantitative phase retrieval in a lensless phase imaging system as thestandard platform and show that the two questions are related and share acommon crux: the choice of the training examples. Moreover, we connect thestrength of the regularization effect imposed by a training set to the trainingprocess with the Shannon entropy of images in the dataset. That is, the higherthe entropy of the training images, the weaker the regularization effect can beimposed. We also discover that weaker regularization effect leads to betterlearning of the underlying propagation model, i.e. the weak object transferfunction, applicable for weakly scattering objects under the weak objectapproximation. Finally, simulation and experimental results show that bettercross-domain generalization performance can be achieved if DNN is trained on ahigher-entropy database, e.g. the ImageNet, than if the same DNN is trained ona lower-entropy database, e.g. MNIST, as the former allows the underlyingphysics model be learned better than the latter.

On the interplay between physical and content priors in deep learning for computational imaging