3D Smoothly Blended Patch-Based Prediction of Extended Volumes

--- DISCLAIMER: REPO IN BROKEN STATE ---

This repo is currently in a BROKEN state and updates are not in short-time schedule as for now. If you're interested in using SP3D you can submit an issue and I'll schedule maintainance.

Performs 3D patch-based, prediction of extended volumes averaging over different flips and rotations of the same volume to reduce noise.

For every view each voxel is predicted n times using sliding 3D windows, border effects are minimized using a 3D second order spline window function to prioritize the central area of the receptive field.

Implementation details to be added soon.

Derivative work of 2D analogous project https://github.com/Vooban/Smoothly-Blend-Image-Patches by Vooban Inc

How to use:

from smooth_predict3D import smooth_predict3D

from your_code import your_model

# Instanciate a U-Net CNN (or any similarly-behaved neural network) in the variable named `model`. We use a Keras model but it can be anything:
model = your_model()

# CNN's receptive field's border size: size of patches
window_size = 160

# Amount of categories predicted per pixels.
nb_classes = 10

# Load an image. Convention is channel_last, such as having an input_img.shape of: (x, y, z, nb_channels), where nb_channels is of 3 for regular RGB images. If your model has a different input scheme pre-transform the data.
input_img = ...

# Use the algorithm, 'pred_func' is used to predict batches of patches, in (batch, x, y, z, nb_channels), if your model has a different input scheme define pred_func as a function to accept (batch, x,y,z, channels) as input and return (batch, x,y,z, predict_classes)

# Note that model.predict(...) accepts a 4D tensor of shape (batch, x, y, z, nb_channels), such as a Keras model.

predictions_smooth = predict_img_with_smooth_windowing(
    input_img,
    window_size=window_size,
    subdivisions=2,  # Minimal amount of overlap for windowing. Must be an even number.
    nb_classes=nb_classes,
    pred_func=(
        lambda img_batch_subdiv: model.predict(image_to_neural_input(img_batch_subdiv))
    ),
    max_batch = 10 # Maximum numbers of patches in a batch
)

# For more details, refer to comments in code

Coded by Filippo Castelli

Forked from original project by Guillaume Chevalier

For more information

If you need anything don't hesitate to open an issue on the repo page!

filippocastelli/smooth_predict3D

3D Smoothly Blended Patch-Based Prediction of Extended Volumes

--- DISCLAIMER: REPO IN BROKEN STATE ---

How to use:

For more information