v1.0 - Pretrained Model Weights for Back-in-Time Diffusion (BTD)

This release includes the pretrained model weights for the Back-in-Time Diffusion (BTD) project, designed for unsupervised detection of medical deepfakes in MRI and CT scans.

Contents:

• MRI_model.pt: Pretrained weights for the MRI model.
• CT_model.pt: Pretrained weights for the CT model.

Usage:

To use these weights in your own project or in a Colab environment, download the MRI_model.pt and CT_model.pt files and load them using the following code snippet:

import torch
from denoising_diffusion_pytorch import Unet, GaussianDiffusion

# Paths to the downloaded weights
CT_weights = "path_to_CT_model.pt"
MRI_weights  = "path_to_MRI_model.pt"

# Initialize the UNet model
unet = Unet(
    dim = 32,
    dim_mults = (1, 2, 4, 8),
    channels = 1
)

# Load the CT model
CT_model = GaussianDiffusion(
    unet,
    objective = "pred_noise",
    image_size = 96,
    timesteps = 1000,
    sampling_timesteps = 250 
).to(device)

CT_model.load_state_dict(torch.load(CT_weights)['model'])
CT_model.eval()

# Load the MRI model
MRI_model = GaussianDiffusion(
    unet,
    objective = "pred_noise",
    image_size = 128,
    timesteps = 1000,
    sampling_timesteps = 250   
).to(device)

MRI_model.load_state_dict(torch.load(MRI_weights)['model'])
MRI_model.eval()