Data and scripts used in: "Reconstructing Quantum Dot Charge Stability Diagrams with Diffusion Models"

This dataset contains the data and scripts used in the paper ''Reconstructing Quantum Dot Charge Stability Diagrams with Diffusion Models''. We train a lightweight diffusion model to reconstruct missing measurements in Charge Stability Diagrams. The models are trained using the Delft-CSD-Dataset, and it is necessary to download the dataset to reproduce our results. Here we show the files and folders contained in this repository, in the order to be run if one wants to reproduce our results. csd_images.zip contains the CSD images from the Delft-CSD-Dataset, converted to grayscale. utils.py and utils_analysis.py contain functions used throughout the training, validation, and assesment stage. diffusion.py contains a class used to perform the diffusion forward step. model.py contains the classes used to build the neural network that performs the denoising/backward step. create_dataset_indices.ipynb is used to visualize samples of the dataset and manually select indices used for the test set, and automatically divide the remaining samples in training and validation set. dataset_indices.h5 is generated by create_dataset_indices.ipynb, and contains the indices of the training, validation, and test set. make_configs.py lets the user selected hyperparameters to be fixed and/or sweeped, and created the folders configs_lc for the line-cut mask case, and configs_reduced for the grid mask case. train.py trains the model in model.py for a given hyperparameter settings defined by a config file, and generated results_lc for the line-cut mask case, and results_reduced for the grid mask, with a corresponding config index. val.py calculates the validation loss for the corresponding config index, and stores it in results_lc for the line-cut mask case, and results_reduced for the grid mask. test.py uses trained models to generate reconstructed CSDs, stored in results_lc for the line-cut mask case, and results_reduced for the grid mask, inside the corresponding config index. run_metrics.py takes the reconstructed CSDs and calculate metrics used to assess the quality of the model. It also runs the metrics for the interpolation baseline methods. Numerical results are saved in the files metrics_summary_lc for the line-cut mask case, and metrics_summary_reduced for the grid mask, while images containing edges/ridges results are stored in visualizations_lc and visualizations_reduced. inference_time.py runs the diffusion models to estimate the inference time taken for reconstruction given specific hyperparameters (including number of diffusion steps and mask type). plots.ipynb generates the plots included in the paper.