search

MiguelMonteiro / CRFasRNNLayer

Conditional Random Fields as Recurrent Neural Networks (Tensorflow)
MIT License
82 stars 20 forks source link

The segmentation results are not better than the post-processing CRF, why? #25

Closed zhangrui0828 closed 3 years ago

zhangrui0828 commented 3 years ago

Hello Migule,

Thanks for your repo!!! I am applying Fettpet's code (which is based on your code, can be used on pytorch) with deeplab-v2 for outdoor scene segmentation and compare it the post-processing CRF. Accoring to the introduction of your paper, crfasrff is much better than the post-processing CRF, but I get the same results of them, which means the segmentaion with crfasrnn is not improved comparing with the results of post-processing CRF. I am confused by this. For the CRF parameter set(theta_alpha, theta_beta, theta_gamma), are there any reqirements? Does crfasrnn have pretrained parameters?

Many thanks Rui

MiguelMonteiro commented 3 years ago

From the conclusion of our paper:

We concluded that the performance differences observed were not statically significant and we provide a discussion as to why this technique does not transfer well from 2D RGB images to 3D multi-modal medical images