conditional random field in semantic segmentation

Are CRF (Conditional Random Fields) still actively used in semantic segmentation tasks or do the current deep neural networks made them unnecessary ? I've seen both of the answers in academic papers and, since it seems quite complicated to implement and infer, I would like to have opinions on them before trying them out.

Thank you

The CRFs are still used for the tasks of image labeling and semantic image segmentation along with the DNNs. In fact, CRFs and DNNs are not self-excluding techniques and a lot of recent publications use both of them.

CRFs are based on probabilistic graphical models, where graph nodes and edges represent random variables, initialized with potential functions. DNN can be used as such potential function:

• Conditional Random Fields Meet Deep Neural Networks for Semantic Segmentation
• Conditional Random Fields as Recurrent Neural Networks
• Brain Tumor Segmentation with Deep Neural Network (Future Work Section)

DCNN may be used for the feature extraction process, which is an essential step in applying CRFs:

• Environmental Microorganism Classification Using Conditional Random Fields and Deep Convolutional Neural Networks
• Conditional Random Field and Deep Feature Learning for Hyperspectral Image Segmentation

There are also toolkits, combining both CRFs and DNNs:

• Direct graphical models C++ library