A single linear layer which does a 1x1 conv

A single linear layer which does a 1x1 conv

Defines a PatchGAN discriminator as in Pix2Pix --> see https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/models/networks.py

Returns: img: 3d array, (z,y,x) or (D, H, W)

an unassuming Transformer block

Source: https://github.com/mkocabas/CoordConv-pytorch/blob/master/CoordConv.py

base GPT config, params common to all GPT versions

To value range 0-1 img: 3D, (z, y, x) or (D, H, W) Returns: img: 3d array, (z,y,x) or (D, H, W)

Source: https://github.com/mkocabas/CoordConv-pytorch/blob/master/CoordConv.py

A vanilla multi-head masked self-attention layer with a projection at the end. It is possible to use torch.nn.MultiheadAttention here but I a

Restrict in value range. value > max = max, value < min = min img: 3D, (z, y, x) or (D, H, W) Returns: img: 3d array, (z,y,x) or (D,

Implement of a progressive growing discriminator with ResidualCoordConv Blocks

To Torch Tensor img: 3D, (z, y, x) or (D, H, W) Returns: img: 3d array, (z,y,x) or (D, H, W)

Wraps an arbitrary object with __len__ and __getitem__ into a pytorch dataset

without codebook, Use CT Only, Use original taming modules

without codebook, Use CT Only

Save a checkpoint every N epochs, instead of Lightning's default that checkpoints based on validation loss.

Takes in semi-embeddings

the full GPT language model, with a context size of block_size

GPT-1 like network roughly 125M params

get specific slice from volume img: 3D, (z, y, x) or (D, H, W) Returns: img: 3d array, (z,y,x) or (D, H, W)

credit to @karpathy: https://github.com/karpathy/deep-vector-quantization/blob/main/model.py (thanks!) Gumbel Softmax trick quantizer Cat

without codebook, Use CT, Xray

without codebook, Use CT, Xray

without codebook, Use CT, Xray

without codebook, Use CT, Xray

without codebook, Use CT, Xray

without encoder/decoder, Use CT, Xray

Net2Net Interface for Class-Conditional Model

note: use with a base_lr of 1.0

Returns: img: 3d array, (z,y,x) or (D, H, W)

Returns: img: 3d array, (z,y,x) or (D, H, W)