Published
Edited
Sep 24, 2020
md`# Webgen Logs`
"Created Imaging_Stuff branch in Github uploaded python autoencoder code"
"Looked into javascript tools for using deep learning. ONNX.js is good for saving saved models in a format that can be read and loaded into browser for inference but cannot handle training. tensorflow.js has slower inference but can handle in-browser training. With the sizes of images, it might make most sense to train models locally, then just upload the model for inference via ONNX.js in the browser."
"If we want to allow in-browser training, probably have to shunt away to an external server. I'm not sure how to handle something general enough to allow people to train models for whatever they are interested in. WOuld need a way to input hyperparameter choices"
"Also where to store patch info. If you want to pull patches from slide X, you should be able to pull all patches that contain tissue. Might have to do something like extract patches for all slides locally --> perform white patch filtering --> save list of good patch coordinates for each slide. Can then do things like shuffle patches and choose a random 40, etc.."
Purpose-built for displays of data
Observable is your go-to platform for exploring data and creating expressive data visualizations. Use reactive JavaScript notebooks for prototyping and a collaborative canvas for visual data exploration and dashboard creation.
