v0.1.7

• Add super-resolution implementation.
• Add conv-block to TRec module.

v0.1.6

Add real-loss to new fc_loss.

v0.1.5

Add new fc_loss:
torch.mean((1 + (c1.abs() - c2.abs())**2) * (2 - dot(c1_unit, c2_unit))

v0.1.4.fixed

Rebase of v0.1.4 on v0.1.3.fixed.

v0.1.4

Apply attenuation function to all bin-levels.

Edit: Do not use this release! Use v0.1.4.fixed.

v0.1.3.fixed

v0.1.3 uses the amplitude directly, we should use log(amplitude) to compute the normalization parameters.

v0.1.3

Normalize with mag_min and mag_max computed over train-data.

Edit: Do not use this release! Use v0.1.3.fixed.

v0.1.2

Combine fc- and real-loss.

v0.1.1

Rework projection approach and Fourier coefficient coordinates, and add masking:

• New default projection has a detector of img_shape length.
• Second option has a detector of length img-diagonal.
• Masking can be used to reduce computation costs during training, by masking out high frequencies.

v0.1.0

First implementation of tomographic reconstruction with Fourier Image Transformers.

• MNIST example