Combiner-GANS

Using GANS architecture to solve how broken objects did look like before break. "Use cases Archaeology".

Pix2Pix:

Model_2 in models builder folder represents pix2pix architecture which gave a good results as explained below

Data:

Gathered data of around 600 images belonging to 9 objects "Carton objects are easy to find and tear".

Results So Far:

Predicted Image "Pix2pix"("300 epochs", "Data_version: 3-10-22"):

The model didn't see the first two images on the left, but saw the last image during training.

It needs more training and more data, but unfortunately I've limited resources to continue training with more images.

Please feel free to help our archeologists as you see fit, They do explore our human history and humanity, we can make it easier for them.

I also tried AutoEnconders but didn't get good results