Datasets:

ngaggion
/

ChronoRoot_nnUNet

text stringlengths 0 1.16k

#######################################################################
Please cite the following paper when using nnU-Net:
Isensee, F., Jaeger, P. F., Kohl, S. A., Petersen, J., & Maier-Hein, K. H. (2021). nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nature methods, 18(2), 203-211.
#######################################################################


This is the configuration used by this training:
Configuration name: 2d
{'data_identifier': 'nnUNetPlans_2d', 'preprocessor_name': 'DefaultPreprocessor', 'batch_size': 12, 'patch_size': [448, 576], 'median_image_size_in_voxels': [2464.0, 3280.0], 'spacing': [1.0, 1.0], 'normalization_schemes': ['ZScoreNormalization'], 'use_mask_for_norm': [False], 'UNet_class_name': 'PlainConvUNet', 'UNet_base_num_features': 32, 'n_conv_per_stage_encoder': [2, 2, 2, 2, 2, 2, 2], 'n_conv_per_stage_decoder': [2, 2, 2, 2, 2, 2], 'num_pool_per_axis': [6, 6], 'pool_op_kernel_sizes': [[1, 1], [2, 2], [2, 2], [2, 2], [2, 2], [2, 2], [2, 2]], 'conv_kernel_sizes': [[3, 3], [3, 3], [3, 3], [3, 3], [3, 3], [3, 3], [3, 3]], 'unet_max_num_features': 512, 'resampling_fn_data': 'resample_data_or_seg_to_shape', 'resampling_fn_seg': 'resample_data_or_seg_to_shape', 'resampling_fn_data_kwargs': {'is_seg': False, 'order': 3, 'order_z': 0, 'force_separate_z': None}, 'resampling_fn_seg_kwargs': {'is_seg': True, 'order': 1, 'order_z': 0, 'force_separate_z': None}, 'resampling_fn_probabilities': 'resample_data_or_seg_to_shape', 'resampling_fn_probabilities_kwargs': {'is_seg': False, 'order': 1, 'order_z': 0, 'force_separate_z': None}, 'batch_dice': True}

These are the global plan.json settings:
{'dataset_name': 'Dataset789_ChronoRoot2', 'plans_name': 'nnUNetPlans', 'original_median_spacing_after_transp': [999.0, 1.0, 1.0], 'original_median_shape_after_transp': [1, 2464, 3280], 'image_reader_writer': 'NaturalImage2DIO', 'transpose_forward': [0, 1, 2], 'transpose_backward': [0, 1, 2], 'experiment_planner_used': 'ExperimentPlanner', 'label_manager': 'LabelManager', 'foreground_intensity_properties_per_channel': {'0': {'max': 225.0, 'mean': 119.90813446044922, 'median': 122.0, 'min': 0.0, 'percentile_00_5': 0.3148415684700012, 'percentile_99_5': 196.0, 'std': 38.823753356933594}}}

2025-01-20 15:33:33.043577: unpacking dataset...
2025-01-20 15:33:33.127604: unpacking done...
2025-01-20 15:33:33.128010: do_dummy_2d_data_aug: False
2025-01-20 15:33:33.130200: Using splits from existing split file: nnUNet_preprocessed/Dataset789_ChronoRoot2/splits_final.json
2025-01-20 15:33:33.130489: The split file contains 5 splits.
2025-01-20 15:33:33.130514: Desired fold for training: 0
2025-01-20 15:33:33.130532: This split has 756 training and 189 validation cases.
2025-01-20 15:33:33.661267: Unable to plot network architecture:
2025-01-20 15:33:33.663022: module 'torch.onnx' has no attribute '_optimize_trace'
2025-01-20 15:33:33.683954:
2025-01-20 15:33:33.684024: Epoch 0
2025-01-20 15:33:33.684111: Current learning rate: 0.01
2025-01-20 15:34:25.702976: train_loss 0.2018
2025-01-20 15:34:25.703284: val_loss 0.0585
2025-01-20 15:34:25.703332: Pseudo dice [np.float32(0.0), np.float32(0.0), np.float32(0.0), np.float32(0.0), np.float32(1e-04), np.float32(0.0)]
2025-01-20 15:34:25.703377: Epoch time: 52.02 s
2025-01-20 15:34:25.703405: Yayy! New best EMA pseudo Dice: 0.0
2025-01-20 15:34:26.294650:
2025-01-20 15:34:26.294980: Epoch 1
2025-01-20 15:34:26.295055: Current learning rate: 0.00999
2025-01-20 15:35:14.646918: train_loss -0.0013
2025-01-20 15:35:14.647027: val_loss -0.0872
2025-01-20 15:35:14.647067: Pseudo dice [np.float32(0.0), np.float32(0.0), np.float32(0.6697), np.float32(0.0), np.float32(0.0067), np.float32(0.0)]
2025-01-20 15:35:14.647120: Epoch time: 48.35 s
2025-01-20 15:35:14.647160: Yayy! New best EMA pseudo Dice: 0.011300000362098217
2025-01-20 15:35:15.423238:
2025-01-20 15:35:15.423292: Epoch 2
2025-01-20 15:35:15.423368: Current learning rate: 0.00998
2025-01-20 15:36:03.020656: train_loss -0.1381
2025-01-20 15:36:03.020907: val_loss -0.1816
2025-01-20 15:36:03.020954: Pseudo dice [np.float32(0.2881), np.float32(0.0), np.float32(0.7474), np.float32(0.0), np.float32(0.4787), np.float32(0.0)]
2025-01-20 15:36:03.021007: Epoch time: 47.6 s
2025-01-20 15:36:03.021029: Yayy! New best EMA pseudo Dice: 0.03539999946951866
2025-01-20 15:36:03.935388:
2025-01-20 15:36:03.935457: Epoch 3
2025-01-20 15:36:03.935537: Current learning rate: 0.00997
2025-01-20 15:36:51.606517: train_loss -0.2352
2025-01-20 15:36:51.606756: val_loss -0.3065
2025-01-20 15:36:51.606802: Pseudo dice [np.float32(0.5379), np.float32(0.3189), np.float32(0.774), np.float32(0.0), np.float32(0.7698), np.float32(0.0)]
2025-01-20 15:36:51.606851: Epoch time: 47.67 s
2025-01-20 15:36:51.606889: Yayy! New best EMA pseudo Dice: 0.07190000265836716
2025-01-20 15:36:52.389135:
2025-01-20 15:36:52.389323: Epoch 4
2025-01-20 15:36:52.389404: Current learning rate: 0.00996
2025-01-20 15:37:40.103456: train_loss -0.3428
2025-01-20 15:37:40.103700: val_loss -0.4076
2025-01-20 15:37:40.103748: Pseudo dice [np.float32(0.5844), np.float32(0.5596), np.float32(0.7762), np.float32(0.1596), np.float32(0.7759), np.float32(0.3504)]
2025-01-20 15:37:40.103804: Epoch time: 47.71 s
2025-01-20 15:37:40.103835: Yayy! New best EMA pseudo Dice: 0.11810000240802765
2025-01-20 15:37:40.897617:
2025-01-20 15:37:40.897854: Epoch 5
2025-01-20 15:37:40.897932: Current learning rate: 0.00995
2025-01-20 15:38:28.587023: train_loss -0.4272
2025-01-20 15:38:28.587277: val_loss -0.4336
2025-01-20 15:38:28.587343: Pseudo dice [np.float32(0.5876), np.float32(0.4889), np.float32(0.7751), np.float32(0.3306), np.float32(0.7703), np.float32(0.5668)]
2025-01-20 15:38:28.587382: Epoch time: 47.69 s
2025-01-20 15:38:28.587403: Yayy! New best EMA pseudo Dice: 0.16500000655651093
2025-01-20 15:38:29.368613:
2025-01-20 15:38:29.369121: Epoch 6
2025-01-20 15:38:29.369184: Current learning rate: 0.00995
2025-01-20 15:39:17.039433: train_loss -0.4698
2025-01-20 15:39:17.039688: val_loss -0.4844
2025-01-20 15:39:17.039737: Pseudo dice [np.float32(0.5719), np.float32(0.6184), np.float32(0.7701), np.float32(0.3576), np.float32(0.8051), np.float32(0.5977)]
2025-01-20 15:39:17.039773: Epoch time: 47.67 s
2025-01-20 15:39:17.039794: Yayy! New best EMA pseudo Dice: 0.21050000190734863
2025-01-20 15:39:17.820793:
2025-01-20 15:39:17.825356: Epoch 7
2025-01-20 15:39:17.825446: Current learning rate: 0.00994
2025-01-20 15:40:05.489944: train_loss -0.5131
2025-01-20 15:40:05.490206: val_loss -0.5391
2025-01-20 15:40:05.490256: Pseudo dice [np.float32(0.6362), np.float32(0.6527), np.float32(0.801), np.float32(0.4168), np.float32(0.7839), np.float32(0.6848)]
2025-01-20 15:40:05.490294: Epoch time: 47.67 s
2025-01-20 15:40:05.490315: Yayy! New best EMA pseudo Dice: 0.2556999921798706
2025-01-20 15:40:06.348648:
2025-01-20 15:40:06.383985: Epoch 8
2025-01-20 15:40:06.384077: Current learning rate: 0.00993
2025-01-20 15:40:54.073986: train_loss -0.5274
2025-01-20 15:40:54.074162: val_loss -0.5236
2025-01-20 15:40:54.074249: Pseudo dice [np.float32(0.6372), np.float32(0.6676), np.float32(0.8114), np.float32(0.5108), np.float32(0.8206), np.float32(0.6261)]
2025-01-20 15:40:54.074297: Epoch time: 47.73 s
2025-01-20 15:40:54.074326: Yayy! New best EMA pseudo Dice: 0.2980000078678131
2025-01-20 15:40:54.935162:
2025-01-20 15:40:54.970317: Epoch 9
2025-01-20 15:40:54.970425: Current learning rate: 0.00992
2025-01-20 15:41:42.661351: train_loss -0.552
2025-01-20 15:41:42.661447: val_loss -0.5817

End of preview. Expand in Dataset Viewer.

ChronoRoot nnUNet Dataset

Dataset Description

Dataset Summary

This dataset contains 693 infrared images of Arabidopsis thaliana seedlings with expert annotations for six distinct structural classes, developed for training the ChronoRoot 2.0 segmentation model. The dataset includes raw infrared images and their corresponding multi-class segmentation masks, formatted according to nnUNet requirements.

Supported Tasks

Image Segmentation: Multi-class segmentation of plant structures
Plant Phenotyping: Analysis of root system architecture and plant development

Classes

The dataset includes annotations for six distinct plant structures:

Main Root (Primary root axis)
Lateral Roots (Secondary root formations)
Seed (Pre- and post-germination structures)
Hypocotyl (Stem region between root-shoot junction and cotyledons)
Leaves (Including both cotyledons and true leaves)
Petiole (Leaf attachment structures)

Data Structure

Raw Images: 3280 x 2464 infrared images
Segmentation Masks: Multi-class masks in .png format
nnUNet Configuration: Standard nnUNet dataset organization

Source Data

Images were captured using the ChronoRoot hardware system, featuring:

Raspberry Pi Camera v2
Infrared lighting (850nm)
Optional long pass IR filters (>830nm)
Controlled growth conditions

Dataset Creation

Annotations

Annotation Tool: ITK-SNAP
Annotators: Expert biologists
Verification: Multi-stage quality control process

Personal and Sensitive Information

This dataset contains no personal or sensitive information.

Additional Information

Licensing Information

The dataset is released under GNU GPL 3.0

Citation Information

If you use this dataset, please cite:

@article{gaggion2021chronoroot, title={ChronoRoot: High-throughput phenotyping by deep segmentation networks reveals novel temporal parameters of plant root system architecture}, author={Gaggion, Nicol{'a}s and Ariel, Federico and Daric, Vladimir and Lambert, Eric and Legendre, Simon and Roul{'e}, Thomas and Camoirano, Alejandra and Milone, Diego H and Crespi, Martin and Blein, Thomas and others}, journal={GigaScience}, volume={10}, number={7}, pages={giab052}, year={2021}, publisher={Oxford University Press} }