| { |
| "schema_version": "1.0", |
| "dataset_version": "1.0.0", |
| "references": { |
| "case_manifest": "case_manifest.json", |
| "sampling": "sampling.json", |
| "folds": "../folds/splits_final.json", |
| "nnunet_dataset": "nnunet/dataset.json", |
| "nnunet_plans_directory": "nnunet/plans" |
| }, |
| "preprocessing": { |
| "raw_height_import": { |
| "delimiter": "semicolon", |
| "decimal_separator": ".", |
| "coordinate_handling": "full_numeric_export_top_left_matched_to_label_shape", |
| "empty_rows_and_columns": "remove_fully_empty", |
| "trailing_nonfinite_rows_and_columns": "remove", |
| "shape_matching": { |
| "anchor": "top_left", |
| "excess_rows_and_columns": "crop_bottom_and_right", |
| "missing_rows_and_columns": "pad_bottom_and_right_with_nonfinite" |
| } |
| }, |
| "nonfinite_inpainting": { |
| "method": "OpenCV Telea", |
| "radius_pixels": 3, |
| "temporary_image_dtype": "uint8", |
| "temporary_scaling": { |
| "method": "percentile_linear_scaling", |
| "lower_percentile": 0.5, |
| "upper_percentile": 99.5, |
| "output_range": [ |
| 0, |
| 255 |
| ], |
| "map_back_to_original_height_range": true |
| }, |
| "restore_original_finite_measurements": true |
| }, |
| "normalized_height_output": { |
| "method": "per_sample_normalization", |
| "normalized_range": [ |
| 0, |
| 1 |
| ], |
| "output_dtype": "uint16", |
| "output_range": [ |
| 0, |
| 65535 |
| ] |
| }, |
| "labels": { |
| "foreground_rule": "nonzero_is_foreground", |
| "downsampling_interpolation": "nearest_neighbour" |
| }, |
| "lateral_downsampling": { |
| "image_method": "nonoverlapping_block_mean", |
| "crop_to_factor_multiple": "bottom_and_right" |
| }, |
| "isotropic_z_binning": { |
| "bin_width_um": "native_lateral_sampling_um_per_pixel multiplied by resolution_factor", |
| "quantization": "floor", |
| "formula": "floor(height_um / bin_width_um) multiplied by bin_width_um" |
| } |
| }, |
| "experiments": { |
| "matched_resolution": { |
| "description": "Independent training and testing at each lateral resolution factor.", |
| "training_input": "continuous_height_at_training_factor", |
| "test_input": "continuous_height_at_training_factor", |
| "train_factors": [ |
| 1, |
| 2, |
| 3, |
| 4, |
| 6, |
| 8, |
| 10, |
| 16, |
| 32 |
| ], |
| "test_factors": "same_as_training_factor", |
| "folds": [ |
| 0, |
| 1, |
| 2, |
| 3, |
| 4 |
| ] |
| }, |
| "fixed_native_model": { |
| "description": "The native-resolution model is evaluated on continuous-height inputs block-averaged at each test factor, bilinearly upsampled to the native grid, and then normalized per sample and converted to uint16.", |
| "evaluation_mode": "inference_only", |
| "train_factor": 1, |
| "model_training_input": "native_height_normalized_per_sample_to_uint16", |
| "test_source_representation": "inpainted_continuous_height_before_normalization", |
| "test_factors": [ |
| 1, |
| 2, |
| 3, |
| 4, |
| 6, |
| 8, |
| 10, |
| 16, |
| 32 |
| ], |
| "downsampling": { |
| "method": "nonoverlapping_block_mean", |
| "factor_multiple_handling": "edge_pad_bottom_and_right" |
| }, |
| "upsampling": { |
| "method": "bilinear_interpolation", |
| "target": "native_grid", |
| "final_crop": "bottom_and_right_to_original_shape" |
| }, |
| "post_upsampling": { |
| "normalization": "per_sample_to_0_1", |
| "output_dtype": "uint16" |
| }, |
| "folds": [ |
| 0, |
| 1, |
| 2, |
| 3, |
| 4 |
| ] |
| }, |
| "isotropic_z_binning": { |
| "description": "Height values are quantized at the physical bin width associated with each lateral resolution factor.", |
| "evaluation_mode": "inference_only", |
| "models": "matched_resolution_models_at_the_same_factor", |
| "model_training_input": "continuous_height", |
| "test_input": "z_binned_height", |
| "factors": [ |
| 1, |
| 2, |
| 3, |
| 4, |
| 6, |
| 8, |
| 10, |
| 16, |
| 32 |
| ], |
| "folds": [ |
| 0, |
| 1, |
| 2, |
| 3, |
| 4 |
| ] |
| }, |
| "detrending": { |
| "description": "A least-squares plane is subtracted from each height map before the native-resolution pipeline.", |
| "method": "least_squares_plane_subtraction", |
| "resolution_factor": 1, |
| "training_input": "detrended_continuous_height", |
| "test_input": "detrended_continuous_height", |
| "folds": [ |
| 0, |
| 1, |
| 2, |
| 3, |
| 4 |
| ] |
| }, |
| "leave_one_papyrus_out": { |
| "description": "Each partition holds out all samples from one papyrus.", |
| "resolution_factor": 1, |
| "training_input": "continuous_height", |
| "test_input": "continuous_height", |
| "folds": [ |
| 5, |
| 6, |
| 7 |
| ], |
| "held_out_papyri_by_fold": { |
| "5": "PHerc. 500P2", |
| "6": "PHerc. 250", |
| "7": "PHerc. 248" |
| } |
| } |
| }, |
| "dataset_mappings": { |
| "continuous_height": [ |
| { |
| "factor": 1, |
| "dataset_id": 701, |
| "dataset_name": "Dataset701_Profilometry2D" |
| }, |
| { |
| "factor": 2, |
| "dataset_id": 702, |
| "dataset_name": "Dataset702_Profilometry2D" |
| }, |
| { |
| "factor": 3, |
| "dataset_id": 703, |
| "dataset_name": "Dataset703_Profilometry2D" |
| }, |
| { |
| "factor": 4, |
| "dataset_id": 704, |
| "dataset_name": "Dataset704_Profilometry2D" |
| }, |
| { |
| "factor": 6, |
| "dataset_id": 706, |
| "dataset_name": "Dataset706_Profilometry2D" |
| }, |
| { |
| "factor": 8, |
| "dataset_id": 708, |
| "dataset_name": "Dataset708_Profilometry2D" |
| }, |
| { |
| "factor": 10, |
| "dataset_id": 710, |
| "dataset_name": "Dataset710_Profilometry2D" |
| }, |
| { |
| "factor": 16, |
| "dataset_id": 716, |
| "dataset_name": "Dataset716_Profilometry2D" |
| }, |
| { |
| "factor": 32, |
| "dataset_id": 732, |
| "dataset_name": "Dataset732_Profilometry2D" |
| } |
| ], |
| "isotropic_z_binned_height": [ |
| { |
| "factor": 1, |
| "dataset_id": 801, |
| "dataset_name": "Dataset801_Profilometry2D_z0p68793625_ds1" |
| }, |
| { |
| "factor": 2, |
| "dataset_id": 802, |
| "dataset_name": "Dataset802_Profilometry2D_z1p3758725_ds2" |
| }, |
| { |
| "factor": 3, |
| "dataset_id": 803, |
| "dataset_name": "Dataset803_Profilometry2D_z2p06380875_ds3" |
| }, |
| { |
| "factor": 4, |
| "dataset_id": 804, |
| "dataset_name": "Dataset804_Profilometry2D_z2p751745_ds4" |
| }, |
| { |
| "factor": 6, |
| "dataset_id": 806, |
| "dataset_name": "Dataset806_Profilometry2D_z4p1276175_ds6" |
| }, |
| { |
| "factor": 8, |
| "dataset_id": 808, |
| "dataset_name": "Dataset808_Profilometry2D_z5p50349_ds8" |
| }, |
| { |
| "factor": 10, |
| "dataset_id": 810, |
| "dataset_name": "Dataset810_Profilometry2D_z6p8793625_ds10" |
| }, |
| { |
| "factor": 16, |
| "dataset_id": 816, |
| "dataset_name": "Dataset816_Profilometry2D_z11p00698_ds16" |
| }, |
| { |
| "factor": 32, |
| "dataset_id": 832, |
| "dataset_name": "Dataset832_Profilometry2D_z22p01396_ds32" |
| } |
| ] |
| }, |
| "training": { |
| "framework": "nnUNetv2", |
| "trainer": "nnUNetTrainer", |
| "configuration": "2d", |
| "plans_identifier": "nnUNetResEncUNetPlans_24G", |
| "epochs": 1000, |
| "training_iterations_per_epoch": 250, |
| "validation_iterations_per_epoch": 50, |
| "early_stopping": false, |
| "checkpoint_for_inference": "checkpoint_best.pth", |
| "loss": "Dice plus cross-entropy", |
| "optimizer": { |
| "name": "SGD", |
| "nesterov": true, |
| "initial_learning_rate": 0.01, |
| "weight_decay": 0.00003, |
| "schedule": "polynomial_decay" |
| }, |
| "foreground_oversampling_fraction": 0.33 |
| }, |
| "software": { |
| "python": "3.12.2", |
| "numpy": "1.26.4", |
| "opencv_python": "4.11.0.86", |
| "pytorch": "2.5.1", |
| "torchvision": "0.20.1", |
| "nnunetv2": "2.5.1", |
| "nnunet_git_commit": "834b80f" |
| } |
| } |
|
|
