Computer Vision News - April 2023

36 Medical Imaging Tools recons_loss = l1_loss(reconstruction.float(), images. float()) val_loss += recons_loss.item() val_loss /= val_step val_recon_epoch_loss_list.append(val_loss) total_time = time.time() - total_start print(f"train completed, total time: {total_time}.") To check the training, you can plot the reconstructed images at every x validation steps (here set to 10), e.g. To write this article, we used the tutorials found on Github , an easily approachable way to get into MONAI Generative Models. For the second experiment on Inpainting, we used a pre-trained model offered by the creators and available here . This is an additional asset of this tool, which will soon offer models (in the MONAI model zoo platform) pre-trained on big datasets ready for fine-tuning. For now, at this link , you will find the weights from two pre-trained models for brain images and chest X-Ray generator. In our previous article onMONAI, we also generated synthetic images of hands, using GANs. Check it out to understand what amazing progress the MONAI team has achieved in just a bit more than 2 years!