-
Insert new photos without wrapped paper
-
Add possibility to update template metadata
-
Sort matches, use the best X matches
-
Select the two regions of interest
-
Calculate transformation between the two images (suppress outliers?)
-
Get list of filled boxes (images!) by using the transformation
-
Run edge detection for each; maybe better use a threshold?
-
Calculate bounding box(es) algorithm on each one
-
Train a neural network on MNIST using PyTorch, also run the bounding box algorithm on each sample