Skip to content
This repository has been archived by the owner on Oct 19, 2024. It is now read-only.

Retrain from scratch

Alexandre Vicenzi edited this page Oct 10, 2019 · 1 revision

You can retrain from scratch by using the training notebooks (ColorizeTrainingArtistic.ipynb, for example).

You need to download whatever dataset you want to use (recommended is ImageNet) and extracting it into the directory the notebook is looking at.

The path in the notebook by default is data/imagenet/ILSVRC/Data/CLS-LOC, the path is relative to where you're running the Jupyter instance from (which we assume to be the same folder as your notebook).

The CLS-LOC folder should contain a train and val folder, and inside those contains a folder for each class of objects, and inside each of those are the original color photos themselves.

Once you have those, then just run the notebook and follow the instructions.

Clone this wiki locally