cat and Dog example only sees cats

good morning guys
I’m trying to enter the DL world…but the result is moderate so far…
the first workflow “preprocess image datadb” has this imagcalc node that is replacing the images…I assume it has to append the calc rather than replacing the image column…?

but somehow the dogs are still not recognised as such… looking into the python node I get the following log:

Using TensorFlow backend.
2018-11-30 09:13:42.200898: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX
C:\Users\User\Anaconda3\envs\py35_knime\lib\site-packages\keras\engine\saving.py:292: UserWarning: No training configuration found in save file: the model was not compiled. Compile it manually.
warnings.warn(‘No training configuration found in save file: ’
C:\Users\User\Anaconda3\envs\py35_knime\lib\site-packages\keras\engine\sequential.py:110: UserWarning: Sequential.model is deprecated. Sequential is a subclass of Model, you can just use your Sequential instance directly.
warnings.warn(’Sequential.model is deprecated. ’

but I have no clue what to do with this… any hints are highly apprechiated

Hi @adaptagis,

Can you validate that the training data (Image+label) you use is correct? The dataset was updated recently. Can you try to attach a sorter next to the ListFiles node and sort by url / path to make sure that the data is in the right order if it wasn’t before? Assumption for training data is that the first 12000(?) rows are of one class, the second 12000 rows dogs.

Thanks,

Christian

SOLVED

hallo Christian Danke für die prompte Antwort!

so the dataset I use for training is much smaler…has only 2025 rows…but I adapted the rule engine and all the absolute partitioning into relative % of the dataset…(I’m not a friend of absolute figures there anyway…)
there I discovered that on the first workflow in the rule engine the row separator was slightly wrong set and two dogs where dedicated to cats as well and I corrected that.
the classifier works well and gives the right class to the images…

but I’m still not sure wether the imCalc node in Wfl1 should really replace image or not… as I don’t see the images at the end of workflow 2 (trasin simple CNNdb)

Hi @adaptagis,
the Image calculator is necessary to normalize the images for the network. You can keep the original images if you want but make sure that only the normalized images are fed to the network.

Cheers,

nemad

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.