Visualizer

The Visualizer component provides a qualitative analysis of the model.
It is possible to visualize the ground truth and the predictions at different levels of granularity.

VisualizerClassification

The VisualizerClassification class can be used to visualize the ground truth, the predictions and the CAMs of classification tasks.

Parameters

dataset: DatasetClassification or DatasetCAMs; Data set used for the visualization
analyzers: list, optional; List of the analyzers of different models.
(default is None)
is_image: bool, optional; Indicates whether the data set represent images or not.
(default is True)
custom_display_function: function, optional; User custom display function. If not specified, it is used the default one.
(default is None)

Example

For ground truth visualization

from odin.classes import DatasetClassification, VisualizerClassification

dataset_gt_param = "/path/to/gt/file.json" # Your file gt file goes here
images_path = "/path/to/gt/images" # Your images folder goes here
classification_type = TaskType.CLASSIFICATION_MULTI_LABEL

my_dataset = DatasetClassification(dataset_gt_param, classification_type,
                                   observations_abs_path=images_path)

my_visualizer = VisualizerClassification(my_dataset)

For ground truth and predictions visualization

from odin.classes import DatasetClassification, VisualizerClassification

dataset_gt_param = "/path/to/gt/file.json"
images_path = "/path/to/gt/images"
path_to_detections = [("model_a", "/path/to/predictions_a"),
                      ("model_b", "/path/to/predictions_b")]
classification_type = TaskType.CLASSIFICATION_MULTI_LABEL

my_dataset = DatasetClassification(dataset_gt_param, classification_type,
                                   proposals_paths=path_to_detections,
                                   observations_abs_path=images_path)

my_visualizer = VisualizerClassification(my_dataset)

For ground truth, predictions and analyses visualization

from odin.classes import AnalyzerClassification, VisualizerClassification

my_analyzer_a = AnalyzerClassification("model_a", my_dataset)
my_analyzer_b = AnalyzerClassification("model_b", my_dataset)

my_visualizer = VisualizerClassification(my_dataset,
                                         analyzers=[my_analyzer_a, my_analyzer_b])

For CAMs visualization

from odin.classes import DatasetCAMs, VisualizerClassification

dataset_gt_param = "/path/to/gt/file.json" # Your file gt file goes here
images_path = "/path/to/gt/images" # Your images folder goes here
path_to_cams_detections = [("model_a", "/path/to/cams/predictions_a"),
                           ("model_b", "/path/to/cams/predictions_b")]
classification_type = TaskType.CLASSIFICATION_MULTI_LABEL


my_dataset = DatasetCAMs(dataset_gt_param, classification_type,
                         cams_paths=path_to_cams_detections,
                         observations_abs_path=images_path,
                         for_analysis=True)

my_visualizer = VisualizerClassification(my_dataset)

VisualizerLocalization

The VisualizerLocalization class can be used to visualize the ground truth and the predictions of localization tasks, such as object detection and instance segmentation.

Parameters

dataset: DatasetLocalization; Data set used for the visualization
analyzers: list, optional; List of the analyzers of different models.
(default is None)

Example

For ground truth visualization

from odin.classes import DatasetLocalization, VisualizerLocalization

dataset_gt_param = "/path/to/gt/file.json" # Your file gt file goes here
images_path = "/path/to/gt/images" # Your images folder goes here
localization_type = TaskType.OBJECT_DETECTION

my_dataset = DatasetLocalization(dataset_gt_param, localization_type,
                                 images_abs_path=images_path)

my_visualizer = VisualizerLocalization(my_dataset)

For ground truth and predictions visualization

from odin.classes import DatasetLocalization, VisualizerLocalization

dataset_gt_param = "/path/to/gt/file.json"
images_path = "/path/to/gt/images"
path_to_detections = [("model_a", "/path/to/predictions_a"),
                      ("model_b", "/path/to/predictions_b")]
localization_type = TaskType.OBJECT_DETECTION

my_dataset = DatasetLocalization(dataset_gt_param, localization_type,
                                 proposals_paths=path_to_detections,
                                 images_abs_path=images_path)

my_visualizer = VisualizerLocalization(my_dataset)

For ground truth, predictions and analyses visualization

from odin.classes import AnalyzerLocalization, VisualizerLocalization

my_analyzer_a = AnalyzerLocalization("model_a", my_dataset)
my_analyzer_b = AnalyzerLocalization("model_b", my_dataset)

my_visualizer = VisualizerLocalization(my_dataset,
                                       analyzers=[my_analyzer_a, my_analyzer_b])