This is a project where I try to predict images' deformations. Each input image with generated text into it is deformed in different ways. And my purpose is to build a tensorflow model which would be capable to predict 8 deformation values(each angle has 2 coordinates so if we shift all angles --> 8 values are changed).
-
Training set has 10000 images
- Image size 400x600
- Each input image has text into it
- The size of text letters ranges from 1 to 3
- The fond of text letters can be one of three types
- All input images are deformed: each angle is shifted inward along the sides of the image.A shift value ranges from 0 to 0.2 of the width and height of the initial image.
- Model and training set generator must be defined in individual files to be readable
- In Data_generator notebook you can see training examples here
- In Model notebook you can see a Tensorflow model with its results on training,validation and test sets here
- Visualised restored images
