Tensorflow-Project-Template

Why we included this project

Starting a new TensorFlow model usually means redoing the same setup work: sessions, savers, data loaders, and logging. This template packages that boilerplate into a folder structure with abstract base classes, so you extend a few classes instead of rewriting them. Checkpoint handling, counters, summary logging, and optional Comet.ml reporting come ready, and you only override two methods to define your model and its training loop. That makes it a useful pattern to copy for an object-oriented TensorFlow 1.x codebase, and a fast way to prototype something like a VGG-style network. The example files are worth studying if you are learning how to organize a deep learning project.