Computer Vision News - April 2023

32 Medical Imaging Tools val_loss = F.mse_loss(noise_pred.float(), noise.float()) val_epoch_loss += val_loss.item() progress_bar.set_postfix({"val_loss": val_epoch_loss / (step + 1)}) val_epoch_loss_list.append(val_epoch_loss / (step + 1)) After training, we tried to sample brain images using as starting point an image containing just noise and this is the result: INPAINTING (USING PRE-TRAINED MODEL) from generative.inferers import DiffusionInferer from generative.networks.nets import DiffusionModelUNet from generative.networks.schedulers import DDPMScheduler Loading MEDNIST Dataset val_transforms = transforms.Compose( [ transforms.LoadImaged(keys=["image"]), transforms.EnsureChannelFirstd(keys=["image"]), transforms.ScaleIntensityRanged(keys=["image"], a_min=0.0, a_ max=255.0, b_min=0.0, b_max=1.0, clip=True), ] ) val_ds = CacheDataset(data=val_datalist, transform=val_transforms) val_loader = DataLoader(val_ds, batch_size=128, shuffle=False, num_work- ers=4, persistent_workers=True) Model: Diffusion Model U-Net (2D) model = DiffusionModelUNet( spatial_dims=2, in_channels=1, out_channels=1, num_channels=(128, 256, 256), attention_levels=(False, True, True),