Coloring greyscale images with deep learning.
For more details on this, Kindly check the paper - https://github.com/baldassarreFe/deep-koalarization/blob/master/paper.pdf
I tried to implement the same architecture as present in the paper, however, due to GPU and RAM limitation, I was unable to complete it.
I removed the fusion of inception network and trained on nearly 5000 images. Which gave average results but not satisfactory.
Key takeaways -
- The network is very bad in generalizing at a lower number of epochs with a varied dataset, maybe I have to try with a larger dataset and more number of epochs.
- I suggest going with a domain-specific dataset, for example, coloring images of trees, mountains etc, train the model with a similar dataset.
Image dataset was downloaded from google images, refer to link (https://www.pyimagesearch.com/2017/12/04/how-to-create-a-deep-learning-dataset-using-google-images/)