Supplemental Data for manuscript entitled
"LeafNet for Segmenting and Quantifying Stomata and Pavement Cells"

Figure 1:

Images

  • F1_training_data.zip [203.9 MB]
    Dataset of 140 training images shown in Figure 1D, which are used to train StomaNet and to evaluate LeafSeg.

Figure 2:

Images

Data tables

  • F2_stoma_count.CSV [719.0 Bytes]
    Stoma counts from ground truth and StomaNet's prediction on F2_validation_data.zip. The data are used for Figure 2B. This data table was generated from the comparison between F2_validation_data.zip and F2_stoma_prediction.zip.
  • F2_stoma_size_error_rate.txt [9.6 KB]
    Error rate of stoma sizes in StomaNet's prediction of F2_validation_data.zip. The data are used for Figure 2E-F. This data table was generated from the comparison between F2_validation_data.zip and F2_stoma_prediction.zip.
  • F2_stoma_length_width_ratio_error_rate.txt [16.0 KB]
    Error rate of stoma length / width ratios in StomaNet's prediction of F2_validation_data. The data are used for Figure 2E,F. This data table was generated from the comparison between F2_validation_data.zip and F2_stoma_prediction.zip.
  • F2_pavement_cell_count.CSV [4.9 KB]
    Pavement cell counts from ground truth and LeafSeg's segmentation of F1_training_data.zip. The data are used for Figure 2D. This data table was generated from the comparison between F1_training_data.zip and F2_pavement_cell_prediction.zip.
  • F2_pavement_cell_sizes_manual.txt [67.5 KB]
    All pavement cell sizes in ground truth of F1_training_data.zip. The data are used in Figure 2G. This data table was generated from F1_training_data.zip.
  • F2_pavement_cell_sizes_predicted.txt [64.9 KB]
    All pavement cell sizes in LeafSeg's segmentation of F1_training_data.zip. The data are used in Figure 2G. This data table was generated from F2_pavement_cell_prediction.zip.

Figure 4

Images

Data tables

  • F4_ITK_cell_count.csv [11.1 KB]
    Pavement cell counts from ground truth and ITK morphological watershed's segmentation on F1_training_data.zip. The data are used in Figure 4A. This data table was generated from the comparison between F1_training_data.zip and F4_ITK_cell_prediction.zip.
  • F4_PlantSeg_cell_count.csv [11.1 KB]
    Pavement cell counts from ground truth and PlantSeg's segmentation of F1_training_data.zip. The data are used in Figure 4A. This data table was generated from the comparison between F1_training_data.zip and F4_PlantSeg_cell_prediction.zip.
  • F4_CellPose_CytoModel_cell_count.csv [3.3 KB]
    Pavement cell counts from ground truth and Cellpose (CytoModel)'s segmentation of F1_training_data.zip. The data are used in Figure 4A. This data table was generated from the comparison between F1_training_data.zip and F4_CellPose_CytoModel_cell_prediction.zip.
  • F4_CellPose_RetrainedModel_cell_count.csv [3.3 KB]
    Pavement cell counts from ground truth and CellPose-Retrained's segmentation of F1_training_data.zip. The data are used in Figure 4A. This data table was generated from the comparison between F1_training_data.zip and F4_CellPose_RetrainedModel_cell_prediction.zip.
  • F4_PaceQuant_cell_count.csv [11.2 KB]
    Pavement cell counts from ground truth and PaCeQuant's segmentation of F1_training_data.zip. The data are used in Figure 4A. This data table was generated from the comparison between F1_training_data.zip and F4_PaceQuant_cell_prediction.zip.

Figure 5

Images

Data tables

  • F5_ABC_LeafNet_pavement_cell_prediction.csv [431.0 Bytes]
    LeafNet's performance on pavement cell segmentation in tobacco images. This data table was generated from the comparison between F5_ABC_ground_truth.zip and F5_ABC_LeafNet_prediction.zip.
  • F5_ABC_LeafNet_stoma_prediction.csv [316.0 Bytes]
    LeafNet's performance on stoma detection in tobacco images. This data table was generated from the comparison between F5_ABC_ground_truth.zip and F5_ABC_LeafNet_prediction.zip.
  • F5_ABC_StomataCounter_stoma_prediction.csv [316.0 Bytes]
    StomataCounter's performance on stoma detection in tobacco images. This data table was generated from the comparison between F5_ABC_ground_truth.zip and F5_ABC_StomataCounter_prediction.zip.
  • F5_FGH_pavement_cell_prediction.csv [301.0 Bytes]
    Performance of different softwares on pavement cell segmentation in confocal images. This data table was generated from the comparison between F5_FGH_ground_truth.zip versus F5_FGH_LeafNet_prediction.zip, F5_FGH_MGX_prediction.zip, and F5_FGH_PlantSeg_prediction.zip.
  • F5_FGH_stoma_prediction.csv [198.0 Bytes]
    Performance of different softwares on stoma detection in confocal images. This data table was generated from the comparison between F5_FGH_ground_truth.zip versus F5_FGH_LeafNet_prediction.zip and F5_FGH_StomataCounter_prediction.zip.

Figure 6

Images

  • F6_ABC_rough_training_data.zip [2.9 GB]
    Dataset of 960 training images roughly labeled by StomataCounter. These images were used to train StomaNet universal.
  • F6_ABC_manual_training_data.zip [386.7 MB]
    Dataset of 140 manually labeled training images. These images were used to train StomaNet universal.
  • F6_ABC_testing_data.zip [124.8 MB]
    Dataset of 47 testing images used in Figure 6. These images were used to evaluate StomaNet universal and StomataCounter.
  • F6_ABC_StomaNet_results.zip [1.4 MB]
    StomaNet's prediction (annotation images) on F6_ABC_testing_data.zip with different thresholds.
  • F6_ABC_StomataCounter_results.zip [183.3 MB]
    StomataCounter's prediction (csv file and heatmap) on F6_ABC_testing_data.zip with different thresholds. Please note that the input images are scaled by 2x, and thus the reported positions should be divided by 2 before using them.
  • F6_DEFGHI_testing_data.zip [654.6 MB]
    Dataset of 223 testing images used in Figure 6. These images were used to evaluate different cell segmentation tools.
  • F6_DEFGHI_LeafSeg_results.zip [298.8 MB]
    LeafSeg's prediction (annotation images) on F6_DEFGHI_testing_data.zip. The threshold was set to 70 for best performance.
  • F6_DEFGHI_Cellpose_results.zip [12.0 MB]
    Cellpose's prediction on F6_DEFGHI_testing_data.zip
  • F6_DEFGHI_ITK_results.zip [15.3 MB]
    ITK morphological watershed's prediction on F6_DEFGHI_testing_data.zip. The parameter level was set to 5 for best performance.
  • F6_DEFGHI_PlantSeg_results.zip [6.9 MB]
    PlantSeg's prediction on F6_DEFGHI_testing_data.zip. The parameter Under-/Over-segmentation factor was set to 0.8, and the CNN prediction threshold was set to 0.25 for best performance.

Data tables

  • F6_ABC_StomaNet_results_collected.zip [3.8 KB]
    Stoma counts from ground truth and StomaNet's prediction of F6_ABC_testing_data.zip The data were used in Figure 6B and Figure 6C. This data table was generated from the comparison between F6_ABC_testing_data.zip and F6_ABC_StomaNet_results.zip.
  • F6_ABC_StomataCounter_results_collected.zip [8.2 KB]
    Stoma counts from ground truth and StomataCounter's prediction of F6_ABC_testing_data.zip. The data were used in Figure 6B and Figure 6C. This data table was generated from the comparison between F6_ABC_testing_data.zip and F6_ABC_StomataCounter_results.zip.
  • F6_DEFGHI_LeafSeg_pavement_cell_prediction.csv [19.1 KB]
    LeafSeg's performance on pavement cell segmentation in F6_DEFGHI_testing_data.zip. This data table was generated from the comparison between F6_DEFGHI_testing_data.zip and F6_DEFGHI_LeafSeg_results.zip.
  • F6_DEFGHI_Cellpose_pavement_cell_prediction.csv [19.2 KB]
    Cellpose's performance on pavement cell segmentation in F6_DEFGHI_testing_data.zip. This data table generated from the comparison between F6_DEFGHI_testing_data.zip and F6_DEFGHI_Cellpose_results.zip.
  • F6_DEFGHI_ITK_pavement_cell_prediction.csv [19.2 KB]
    ITK morphological watershed's performance on pavement cell segmentation in F6_DEFGHI_testing_data.zip. This data table was generated from the comparison between F6_DEFGHI_testing_data.zip and F6_DEFGHI_ITK_results.zip.
  • F6_DEFGHI_PlantSeg_pavement_cell_prediction.csv [19.2 KB]
    PlantSeg's performance on pavement cell segmentation in F6_DEFGHI_testing_data.zip. This data table was generated from the comparison between F6_DEFGHI_testing_data.zip and F6_DEFGHI_PlantSeg_results.zip.

Figure 7

Images

LeafNet generated text files

Data tables

  • F7_ABCD_count.csv [10.7 KB]
    LeafNet predicted and manually corrected stoma and pavement cell counts of F7_ABCD_raw_data.zip. The data are used for Figure 7A-D. This data table was generated from F7_ABCD_statistic_data.zip, and manual corrections were done with F7_ABCD_visualized_segmentation.zip.
  • F7_EF_M1_count.csv [635.0 Bytes]
    LeafNet predicted and manually corrected stoma and pavement cell counts of F7_EF_raw_data.zip. The data are used in Figure 7E-F. This data table was generated from F7_EF_statistic_data.zip, and manual corrections were done with F7_EF_visualized_segmentation.zip.
  • F7_EF_M2_count.csv [621.0 Bytes]
    LeafNet predicted and manually corrected stoma and pavement cell counts of F7_EF_raw_data.zip. The data are used in Figure 7E-F. This data table was generated from F7_EF_statistic_data.zip, and manual corrections were done with F7_EF_visualized_segmentation.zip.

Supplemental Figure 9

Images

  • SF9_AD_Denoised_images.zip [79.1 MB]
    Dataset of 140 images preprocessed by StainedDenoiser. The original images are from F1_training_data.zip, in which 0-99 were used for training and validation, and 100-139 were used for evaluation.
  • SF9_AD_CNNwall_prediction.zip [19.8 MB]
    CNNwall's prediction on evaluation images in SF9_AD_Denoised_images.zip. The stomata were masked out.
  • SF9_AD_LeafSeg_results.zip [2.8 MB]
    LeafSeg's prediction (annotation images) on SF9_AD_CNNwall_prediction.zip. The threshold was set to 60 for best performance, and the stomata were masked out.
  • SF9_AD_PlantSeg_results.zip [2.7 MB]
    PlantSeg's prediction with CNNwall model on SF9_AD_Denoised_images.zip. The parameter Under-/Over-segmentation factor was set to 0.5, and the CNN prediction threshold was set to 0.5 for best performance.
  • SF9_BE_Denoised_images.zip [152.1 MB]
    Dataset of 223 images preprocessed by StainedDenoiser. The original images are from F6_DEFGHI_testing_data.zip.
  • SF9_BE_CNNwall_prediction.zip [146.6 MB]
    CNNwall's prediction on images in SF9_BE_Denoised_images.zip.
  • SF9_BE_LeafSeg_results.zip [289.3 MB]
    LeafSeg's prediction (annotation images) on SF9_BE_CNNwall_prediction.zip. The threshold was set to 50 for best performance, and the stomata were masked out.
  • SF9_BE_PlantSeg_results.zip [14.4 MB]
    PlantSeg's prediction with CNNwall model on SF9_BE_CNNwall_prediction.zip. The parameter Under-/Over-segmentation factor was set to 0.8, and the CNN prediction threshold was set to 0.1 for best performance.

Data tables

  • SF9_AD_LeafSeg_pavement_cell_prediction.csv [3.9 KB]
    LeafSeg's performance on pavement cell segmentation in SF9_AD_CNNwall_prediction.zip. This data table was generated from the comparison between F1_training_data.zip and SF9_AD_LeafSeg_results.zip.
  • SF9_AD_PlantSeg_pavement_cell_prediction.csv [3.9 KB]
    PlantSeg's performance with CNNwall on pavement cell segmentation in SF9_AD_Denoised_images.zip. This data table was generated from the comparison between F1_training_data.zip and SF9_AD_PlantSeg_results.zip.
  • SF9_BE_LeafSeg_pavement_cell_prediction.csv [20.9 KB]
    LeafSeg's performance on pavement cell segmentation in SF9_BE_CNNwall_prediction.zip. This data table was generated from the comparison between F6_DEFGHI_testing_data.zip and SF9_BE_LeafSeg_results.zip.
  • SF9_BE_PlantSeg_pavement_cell_prediction.csv [21.0 KB]
    PlantSeg's performance with CNNwall on pavement cell segmentation in SF9_BE_Denoised_images.zip. This data table was generated from the comparison between F6_DEFGHI_testing_data.zip and SF9_BE_PlantSeg_results.zip.