[TL;DR] Join me next week for a webinar on image processing and machine learning with Dash!
Hi everyone! Together with my teammates, we’ve written a couple of image processing apps with Dash, and I’d like to share them with you. You can find the apps and code on the Dash Gallery.
These image processing apps have been developed thanks to a grant of the Chan-Zuckerberg Initiative, who is funding essential software packages for science, including plotly and Dash. Our apps are therefore particularly appropriate for life or material sciences, but are also relevant for any domain of data science and business working with images. In particular, since machine learning is used extensively for image processing, user interaction with image datasets is required at several steps of the machine learning workflow: building a training set thanks to user annotations, and post-processing and exploring the output of the estimator.
Our machine-learning interactive segmentation app allows you to quickly draw freehand annotations over different types of regions which you want to segment. A classifier is trained by the app using these annotations, and the Dash app provides UI elements such as sliders and checklists in order to select the most appropriate features to compute (you can start with the default values which should give a good result). View the code.
The 3D Image Annotation app addresses a tricky problem in 3-D medical imaging modalities such as MRI or Computed Tomography: how to annotate datasets with 2-D annotations when the objects of interest are intrinsically three-dimensional? We have solved this conundrum by pre-computing a so-called “super-pixel” partitioning of the dataset into regions of similar intensity. Then, user-drawn annotations on 2-D slices automatically group together the 3-D superpixels. Check out the code.
These two apps combine the power of Dash for rich and beautiful interactive UIs, and packages of the scientific Python ecosystem such as scikit-image and scikit-learn.
I will be hosting a webinar on November 18th at 2pm EST showcasing how I built these apps, how we developed image annotations for plotly and Dash, as well as some techniques for using scikit-image and scikit-learn to build ML training sets or seeding segmentation algorithms. Hope you can join! https://go.plotly.com/dash-image-processing