search

fiji / Trainable_Segmentation

Fiji library to perform image segmentation based on the Weka learning schemes
https://imagej.net/Trainable_Weka_Segmentation
GNU General Public License v3.0
106 stars 60 forks source link

Explicit definition of sigma values for morphological filters #50

Open acitcratna opened 4 years ago

acitcratna commented 4 years ago

TL;DR: Feature checkboxes for WEKA attributes aren't enough

TL;DR (+1): I can use WEKA to filter and remove features/attributes I don't need for classification, but I can't prevent TWS from generating those features/attributes because I only get to choose a range of sigma values. I have a lot of data, and It's wasting a lot of compute time.

I need more control over sigma values than a range between min/max sigma. I don't fully expect any of the following requests to happen, but I'll never give up hope:

1) Please make it possible to explicitly define explicit sigma values as a list of discrete numbers instead of a range (between minimum and maximum).

2) Even better, please enable explicit definition of sigma values for each individual morphological filter. If I need StdDev values ranging from sigma 1 to 4, I shouldn't have to calculate Difference of Gaussians with sigmas of 1-2, 1-3, 2-3, 3-4... or derivatives at sigmas 1, 2, 3 & 4 when I need it at 8.

Please PLEASE let me define explicit values for sigma

ctrueden commented 4 years ago

@acitcratna Thank you for the suggestions!

I don't fully expect any of the following requests to happen, but I'll never give up hope

Since this is open-source software, you can do better than hope: you (or your group/colleagues/etc.) can improve the plugin yourself and file a pull request here on GitHub! It would be warmly received.

@iarganda Is there anyone in your group able to explore making improvements like this?

@maarzt What is the status of Labkit? Would it address @acitcratna's concerns already?

maarzt commented 4 years ago

@maarzt What is the status of Labkit? Would it address @acitcratna's concerns already?

Yes, Labkit addresses this. It allows to specify sigmas individually for each feature.

ctrueden commented 4 years ago

@maarzt Awesome!

So then: how should we advertise Labkit to the community more prominently? Why not start by editing the "Easy" section of https://imagej.net/Segmentation to feature Labkit over TWS?