Quick setup demoing a project structure that could work well for our project.
npm install
entrypoint (argument parsing, etc)
src/index.js
server-side model / data processing:
src/tools/objprocessor.js
express server (three.js visualization)
src/visualization-server/viz-server.js
web client (three.js visualization)
src/visualization-client/index.js
server-side tensorflow training (TBD)
src/tensorflow-server/tf-server.js
Theoretically the express server + tensorflow server would communicate somehow. tools/ would handle data processing, and would include both functions to do data processing both on individual files and directories / obj globs (in which case it would spin up nodejs clusters / thread workers), and utility functions that could be imported by the tensorflow server and client-side renderer to do various things.
Most sub-components would be capable of spinning up concurrent processes / threads: child processes (viz-server), clusters / thread workers (tools, tensorflows), though tools could -also- operate in a lightweight manner (ie. for batch processing by some other script) when just passed in single files.
This actually works (just doesn't do anything, haha).
do processing on a single file / for testing:
npm run process
=> generates data/params/test-model.json
from data/models/test-model.obj
npm run reconstruct
=> generates data/reconstructed-models/test-model.obj
from data/params/test-model.json
visualize this data
npm run visualize
visualize / browse through all source / reconstructed / generated data
npm run visualize-all
train tensorflow model
npm run tf-train
generate data using current tensorflow model (no training)
npm run tf-generate
visualize tensorflow model while training
npm run visualize-training
visualize generated data from current tensorflow models (no training)
npm run visualize-generation
I went a bit nuts with options and stuff, but at least some of this is implementable.