Skip to content

README.md

Latest commit

 

History

History
34 lines (19 loc) · 1.21 KB

README.md

File metadata and controls

34 lines (19 loc) · 1.21 KB

Colorizing Balck & White Images Using Generative Adverserial Networks

Implemeted a Generational Adverserial Network for colorizing black & white images.
Dataset used : COCO-2017 training
Original paper : Image-to-Image Translation with Conditional Adversarial Networks


Some examples

Things I Tried:

  • Pre-trained the Generator separately:
    • Used U-Net with a pretrained (on ImageNet) Resnet model as its backbone and then pretrained the whole generator model on the dataset.
    • Tried with Resnet18 and Resnet34 models.
  • Tried different Image Augmentations: both train-time (from Albumentation library) and test-time (from TTA library)
  • Tried different optimisers for Generator and Discriminator.

Things to try in Future:

  • Training for more epochs.
  • Hyper-parameter optimisations
  • Create a web application and deploy the model over it.