Autocontext error: object has no attribute

Hi all and especially @k-dominik,

I’m running Autocontext on a series of files batch-extracted with a Fiji macro so I can then use the probability files with object classification. These are .tif stacks (volumes, 512x512, variable number of slices ranging from 34 to 96) . Autocontext appears to work well on all files but one which it refuses to load, the file with only 34 slices. I get the following message: “AutocontextTwoStage” object has no attribute “featureSelectionApplet”.

Here is the file:Nuclei-HRAS-GFP+2-DG_EdU_4_fish1_30012020 - Series002.tif (8.5 MB)

What does this error message mean?

Sincerely,

Matthieu

Hi @MatthieuV,

I assume that you have also larger 3D filters active in your project (so you’ve selected larger sigmas in the feature selection step). Larger sigmas will fail with data that doesn’t “fit” the whole kernel. Your image with only 34 slices is not “deep” enough.

Solutions to this may be:

  1. go back to the project, and compute the sigma 10 features only in 2D (by clicking on the 3D icon). You’d have to confirm that the pixel classification result is still good enough, or add more annotations.
  2. exclude this data from prediction

I assume 2) is not an option ;). In any case, make sure that you check your predictions after changing the filter settings (by enabling live-update in the training).

the error message is… not very helpful :confused:

In order to dodge problems like this in the future, you might include the most narrow image stack in your training data. Then you’ll get a more descriptive error in the gui.

Cheers
Dominik

1 Like