Segmentation visualization issues

[AlbertvanHouten]

Describe the bug

There are a few minor issues in the segmentation visualization/result logic:

• The instance segmentation scene picks a random label color for every scene it draws. This is inconsistent with the other visualizers, which use static colors. Please update https://github.com/open-edge-platform/model_api/blob/master/src/python/model_api/visualizer/scene/segmentation/instance_segmentation.py#L23 to no longer pick a random color. The random color gives ugly results when iterating over a video and every single frame has a different label color. It would also be nice to have an option to display the label name as part of the visualization.
• InstanceSegmentationResult has optional parameter saliency_map, however, when creating an InstanceSegmentationScene and calling the render() function the saliency_map parameter is actually required here: https://github.com/open-edge-platform/model_api/blob/master/src/python/model_api/visualizer/scene/segmentation/instance_segmentation.py#L57
• The same problem also appears for SegmentationResult/Scene
• InstanceSegmentionResult subclasses DetectionResult. This seems unintended and it should subclass Result instead.
• Consider adding a factory for visualization. An example can be found here https://github.com/intel-innersource/applications.ai.geti.geti-edge/blob/main/app/utils/visualization.py

[sovrasov]

@AlbertvanHouten thanks for suggestions!
I see the factory can improve API interactions from user's perspective:

model = Model.create_model(model_path, preload=True)
prediciton = model(img)
overlayed_img = Visuzlizer.overlay_predictions(img, prediciton)

Currently, we have similar interface, but it can not render/overlay:

model = Model.create_model(model_path, preload=True)
visualizer = Visualizer()
visualizer.save(image=img, result=result, path="./vis.jpg")

I think it'd be enough to extend it with render() method, would that be sufficient for your usecase?

[sovrasov]

Please have a look at #334