Plotting¶

Quick guide to plotting tasks in SkiNet.

Notebook with Docker examples: SkiNet/Plotting/docker_plotting_examples.ipynb

Logging¶

Log skipped images to stdout and a file:

import logging
from SkiNet.Utils.loggers import file_logging, stdout_logging
stdout_logging(logging.DEBUG)
file_logging()

Plot images and masks side-by-side¶

From a config-based dataset (ISIC 2017)¶

plot_images_masks_side_by_side_matplotlib(dataset, num_samples=5) indexes dataset[i] for i in range(num_samples) and renders each image and mask with a colorbar. Pass any split from the factory (datasets.train, datasets.val, datasets.test):

from SkiNet.ML.configs.load_config_from_yaml import load_config_from_yaml
from SkiNet.ML.datasets.dataset_factory import create_segmentation_datasets_from_config
from SkiNet.Plotting.plot_images_masks_side_by_side import plot_images_masks_side_by_side_matplotlib

cfg = load_config_from_yaml("main_config.yaml")
datasets = create_segmentation_datasets_from_config(cfg)
plot_images_masks_side_by_side_matplotlib(datasets.val, num_samples=3)

The transforms applied to each split are fixed by get_transform_from_config (train = augmented, val/test = deterministic); the plotter renders whatever the dataset’s transform produces.

Plot images overlaid with masks¶

From a folder using glob patterns¶

folder mode requires both search_pattern_images and search_pattern_masks (a missing or None mask pattern raises ValueError). Patterns are passed to Path(data_root).rglob(...), and filter_and_pair_valid_paths drops unpaired files and pairs whose image/mask sizes differ. Point data_root at the parent holding both the *_Data and *_Part1_GroundTruth directories:

from SkiNet.Plotting.plot_segmentations import plot_segmentations

plot_segmentations(
    mode="folder",
    data_root="/mnt/data",
    search_pattern_images="ISIC-2017_*_Data/*/*.jpg",
    search_pattern_masks="ISIC-2017_*_Part1_GroundTruth/*/*_segmentation.png",
    max_cols=5,
    max_images_to_plot=10,
    alpha=0.5,
)

Dataloader mode (retired)¶

plot_segmentations(mode="dataloader", ...) is retired — it depended on the removed DatasetSplitter and on DatasetClass(data_root=...), a signature only the legacy PH2Dataset supported, and now raises NotImplementedError. To overlay masks for ISIC 2017, build the dataset from config and use the side-by-side or augmented-data plotters below.

Plot augmented data¶

from SkiNet.ML.configs.load_config_from_yaml import load_config_from_yaml
from SkiNet.ML.transformations.transform_data import get_transform_from_config
from SkiNet.ML.datasets.dataset_factory import create_segmentation_datasets_from_config
from SkiNet.ML.transformations.plot_transformed_data import visualize_augmented_data

cfg = load_config_from_yaml("main_config.yaml")
transform = get_transform_from_config(cfg)
datasets = create_segmentation_datasets_from_config(cfg)

visualize_augmented_data(dataset=datasets.train, samples=20)

Saves individual images and a grid of augmented samples to a specified directory.