search

deepmedic / dense3dCrf

Fully-connected (dense) 3D CRF for processing biomedical scans
Other
138 stars 48 forks source link

Calculate CRF parameters #2

Closed parvezamu closed 6 years ago

parvezamu commented 6 years ago

How to calculate CRF parameters

Kamnitsask commented 6 years ago

Grid search I am afraid. The way I did it (and I believe most of the people do the same) is by writing a script that runs this for a few days on some validation data, with random parameters. For each run, you calculate the metric of your preference. After a few trials (days?) you chose the parameters that gave the best results.

Hope this helps. Kostas

parvezamu commented 6 years ago

Thanking You Regards Parvez Ahmad

On Sun, Dec 10, 2017 at 3:25 AM, Konstantinos Kamnitsas < notifications@github.com> wrote:

Grid search I am afraid. The way I did it (and I believe most of the people do the same) is by writing a script that runs this for a few days on some validation data, with random parameters. For each run, you calculate the metric of your preference. After a few trials (days?) you chose the parameters that gave the best results.

Hope this helps. Kostas

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/Kamnitsask/dense3dCrf/issues/2#issuecomment-350499673, or mute the thread https://github.com/notifications/unsubscribe-auth/AWl__JMGYSfBS7N8WvIgAl_bc_RjKOtoks5s-t6tgaJpZM4QxxNZ .

parvezamu commented 6 years ago

Respected Professor I Parvez Ahmad, a new research student wants to take your knowledgeable and valuable suggestions. What will be the use If we bring the uncertainty estimates in the prediction of deep learning models?.Will it help us to improve the inference step ?.Will it be help us to improve the confidence of our overall model's accuracy. Is it beneficial for us to merge together the deep learning and Bayesian machine learning if we use this for example in the segmentation of brain tumor? Kindly provide your valuable suggestions to me. It will be very helpful for me. Expecting the positive response from your side. Thanking You Regards Parvez Ahmad