Runtime Helper

class vortex.runtime.helper.InferenceHelper(model)
__call__(*args, **kwargs)

Run and visualize

Returns

dictionary containing ‘prediction’ and optionally ‘visualization’

Return type

dict

class Visual(class_names)

Helper class for to load images, inference, and visualization for convinience

classmethod draw(result: dict, vis: numpy.ndarray, class_names=None, color_map=None)

draw single prediction result on vis

Parameters

  • result (dict) – single prediction result

  • vis (np.ndarray) – array in which prediction result is to be visualized

  • class_names (mapping, optional) – mapping from int label to human-readable str. Defaults to None.

  • color_map (mapping, optional) – mapping from int to colors. Defaults to colors.

Returns

array with visualiazation

Return type

np.ndarray

classmethod draw_bbox(vis: numpy.ndarray, tl: Tuple[int, int], rb: Tuple[int, int], color: Tuple[int, int, int] = (255, 0, 0))

draw bounding box on vis

Parameters

  • vis (np.ndarray) – [input output] array in which bounding box to be visualized

  • tl (Tuple[int,int]) – top-left

  • rb (Tuple[int,int]) – bottom-right

  • color (Tuple[int,int,int], optional) – desired color of the rectangle. Defaults to colors[0].

Returns

image with bbox visualized

Return type

np.ndarray

classmethod draw_bboxes(vis: numpy.ndarray, bboxes, classes, confidences, color_map=None, class_names=None)

draw multiple bounding box on vis

Parameters

  • vis (np.ndarray) – array in which bounding box is to be visualized

  • bboxes (iterable of tuple) – bounding boxes to be visualized

  • classes (iterable of int) – list of classes corresponding to each bounding box

  • confidences (iterable of float) – list of confidences corresponding to each bounding box

  • color_map (mapping, optional) – mapping from class to color. Defaults to None.

  • class_names (mapping, optional) – mapping from class to str represnting human-readable class name. Defaults to None.

Returns

array with visualization

Return type

np.ndarray

classmethod draw_label(vis, obj_class, confidence, bl, color, class_names=None)

draw single label on vis

Parameters

  • vis (np.ndarray) – array in which label is to be visualized

  • obj_class (integer) – object class/label

  • confidence (scalar) – object confidence

  • bl (tuple) – bottom-left point to visualize label

  • color (tuple) – desired color to visualize label

  • class_names (mapping, optional) – mapping from int label to human-readable str. Defaults to None.

Returns

array with visualiazation

Return type

np.ndarray

classmethod draw_labels(vis, obj_classes, confidences, bls: Sequence[Tuple[int, int]], color_map=None, class_names=None)

draw multiple labels on vis

Parameters

  • vis (np.ndarray) – array in which label is to be visualized

  • obj_classes (iterable) – object classes/labels

  • confidences (iterable) – object confidences

  • bls (Sequence[Tuple[int,int]]) – bottom-left points corresponding to each label to be visualized

  • color_map (mapping, optional) – mapping from int to colors. Defaults to colors.

  • class_names (mapping, optional) – mapping from int label to human-readable str. Defaults to None.

Returns

array with visualiazation

Return type

np.ndarray

classmethod draw_landmarks(vis: numpy.ndarray, landmarks, color: Optional[Tuple[int, int, int]] = None, radius=2, thickness=- 1)

draw multiple landmarks on vis

Parameters

  • vis (np.ndarray) – array in which landmarks are to be visualized

  • landmarks (np.ndarray) – landmarks to be visualized

  • color (Tuple[int,int,int], optional) – desired color of point for visualization. Defaults to None.

  • radius (int, optional) – desired radius of point for visualization. Defaults to 2.

  • thickness (int, optional) – desired thickness of point for visualization. Defaults to -1.

Returns

array with visualization

Return type

np.ndarray

classmethod visualize(batch_vis: List, batch_results: List, class_names=None) → List

draw batched prediction result on vis

Parameters

  • batch_vis (List) – batch image for visualization

  • batch_results (List) – batched prediction result

  • class_names (mapping, optional) – mapping from int label to human-readable str. Defaults to None.

Returns

array with visualiazation

Return type

np.ndarray

classmethod visualize_result(vis: numpy.ndarray, results: List[Dict[str, numpy.ndarray]], class_names=None, color_map=None)

draw single-batch prediction result on vis

Parameters

  • vis (np.ndarray) – array in which prediction result is to be visualized

  • results (List[Dict[str,np.ndarray]]) – single-batch prediction result

  • class_names (mapping, optional) – mapping from int label to human-readable str. Defaults to None.

  • color_map (mapping, optional) – mapping from int to colors. Defaults to colors.

Returns

array with visualiazation

Return type

np.ndarray

classmethod adjust_coordinates(batch_vis, batch_results, coordinate_fmt='relative')

adjust prediction results for visualization

Parameters

  • batch_vis (list) – list of image for visualization

  • batch_results (list) – prediction results to be transformed

  • coordinate_fmt (str, optional) – output coordinat format. Defaults to ‘relative’.

Returns

list of transformed prediction results

Return type

list

static create_runtime_model(**kwargs)

helper method to instantiate model

Returns

wrapped model

Return type

InferenceHelper

classmethod load_images(images)

load images from list of files

Parameters

images (list) – list of files

Raises

TypeError – unknown type passed

Returns

list of loaded np.ndarray

Return type

list

classmethod run_and_visualize(model, images: Union[List[str], numpy.ndarray], output_coordinate_format: str = 'relative', visualize: bool = False, dump_visual: bool = False, output_dir: Union[str, pathlib.Path] = '.', class_names=None, visual=None, **kwargs) → dict

run inference on model with given images paths

Parameters

  • model (vrt.BaseRuntime) – vorted rt model

  • images (Union[List[str],np.ndarray]) – list of image’s path

  • output_coordinate_format (str, optional) – output coordinate format. Defaults to ‘relative’.

  • visualize (bool, optional) – visualize output. Defaults to False.

  • dump_visual (bool, optional) – save images. Defaults to False.

  • output_dir (Union[str,Path], optional) – output directory. Defaults to ‘.’.

Returns

prediction results

Return type

dict

classmethod run_inference(model, batch_imgs, **kwargs)

run inference on batched (possibly non-uniform size) image

Parameters

  • model (runtime) – vortex rt model

  • batch_imgs (list) – list image for inference

Returns

list of dictionary corresponding to each image

Return type

list of dict

classmethod save_images(filenames: List, batch_vis: List, output_dir: Union[str, pathlib.Path] = '.', output_file_prefix='prediction')

save images

Parameters

  • filenames (List) – original filenames

  • batch_vis (List) – visualization results to be saved

  • output_dir (Union[str,Path], optional) – output directory. Defaults to ‘.’.

  • output_file_prefix (str, optional) – prefix to be added to output filenames. Defaults to ‘prediction’.

Returns

output filenames

Return type

list