Improve training

[CameronGBell]

Improves and fleshes out trainable segmentation interface:

• allows up to 5 labels in SegUI
• Updates parameters for trainable segmentation
• enables multi-image training
• adds Filter kernel customisation in SegUI

[codecov[bot]]

Codecov Report

Merging #71 (b0d17ca) into master (5225e33) will increase coverage by 2.10%. The diff coverage is 94.18%.

@@            Coverage Diff             @@
##           master      #71      +/-   ##
==========================================
+ Coverage   86.39%   88.49%   +2.10%     
==========================================
  Files          15       15              
  Lines         992     1156     +164     
==========================================
+ Hits          857     1023     +166     
+ Misses        135      133       -2     

Impacted Files	Coverage Δ
ParticleSpy/segimgs.py	91.57% <91.07%> (+11.81%)	:arrow_up:
ParticleSpy/ParticleAnalysis.py	94.22% <100.00%> (+1.33%)	:arrow_up:
ParticleSpy/custom_kernels.py	100.00% <100.00%> (ø)
ParticleSpy/segptcls.py	94.56% <100.00%> (+0.05%)	:arrow_up:
ParticleSpy/tests/test_particle_clustering.py	100.00% <100.00%> (ø)

[TomSlater]

I think you'd modified the canvas_size to be larger (1024 vs 512) but if you're using a small screen resolution it's then not possible to see the bottom of the window. I've modified to vary the canvas_size with screen resolution. Maybe canvas_size should be the smaller of 0.8*screen height or 1024?

What do you think?

In the long term it would be great to make segUI resizeable.

[CameronGBell]

I think you'd modified the canvas_size to be larger (1024 vs 512) but if you're using a small screen resolution it's then not possible to see the bottom of the window. I've modified to vary the canvas_size with screen resolution. Maybe canvas_size should be the smaller of 0.8*screen height or 1024?

What do you think?

In the long term it would be great to make segUI resizeable.

I think that makes sense, a resizable segUI would make sense to implement alongside a zoomable canvas

[TomSlater]

Okay, I added an if statement to set to 1024 if smaller than the calculated height. I'll leave a resizeable window to you for another PR.

Everything else seems to work for me, so all looks okay. I will go through and change everything to lower case in a PR next week.

One last thing before merging. Could you add an example of training the classifier using an imported mask and not using the segUI? Just for those who don't want to run segUI (for those using cloud-based services etc).