Hello guys you can try the same code using this COLOI dataset. It is simplified dataset made from famous MIRFLICKR-25k and is already converted into LAB color space. So implementing that dataset for the same code shown on the floydhub would be much easier. I would suggest try using Kaggle kernals as you won’t even have to download the dataset, just import it there itself; I’ve also uploaded the weights .h5 file for Inception Resnet V2 in the dataset, so importing that would also become easy.
Source: Deep Learning on Medium