Image Handling - Plot an Image
Is it possible to plot an image to the Results view, especially in the context of the Image Handling Extension? I know that there is an option to Store the Image to a file system, but my ask is to plot the image within the process > Results.
My use cases:
- Teaching to show an image is just a dataset. This is particularly helpful when extending the concept of data shapes outside of tabular 2D datasets.
- After fitting an image classification model, it would be nice to be able to review the predicted labels (e.g. MNIST) with the actual image classified for visual inspection. Perhaps a 3 was misclassified as a 9, and this feature would allow us to visually review how these networks learn (or fail).
My use cases:
- Teaching to show an image is just a dataset. This is particularly helpful when extending the concept of data shapes outside of tabular 2D datasets.
- After fitting an image classification model, it would be nice to be able to review the predicted labels (e.g. MNIST) with the actual image classified for visual inspection. Perhaps a 3 was misclassified as a 9, and this feature would allow us to visually review how these networks learn (or fail).
0
Best Answer
-
ceaperez
MemberPosts:455
Unicorn
Hi@btibert,
please find attached an image of sample process for MNIST dataset.
1- Inside to the Pre-Process Image operator you can create a flows with operators for image preprocessing: crop, rotate, flip, resize.2- To be able to compare the the predicted label and the current label, I use to joint the Exa output of the Perform Operator with a copy of the outuput of the Read Image Metadata Operator used for testing dataset.I hope you find this flow clearest than the previous one.
Best,
Cesar
1


Guru
Answers
Best,
Cesar
Also, when using Read Image As ExampleSet, it doesn't appear that we can do any image pre-processing unless I am missing a Tensor to ExampleSet conversion operator. For example my images need to be rotated.
Thanks again. I am trying to map how I do certain tasks in python within RM (either base or Extensions) and avoid using Execute Python as much as possible.