Visualizer: draw_and_connect_keypoints on more than one instance

Nobody has claimed this yet.

I faced an issue when estimating the skeleton of multiple objects when using draw_and_connect_keypoints it expects the input to belong to one instance only. Although that draw_instance_predictions works on multiple instances. However, it makes plotting of bounding boxes which are not needed sometimes when doing Pose Estimation.

The feature would be useful to add a small level of abstraction that helps people to get involved more into Detectron2. When I tried the function for the first time I didn't know that it accepts only one instance while a predictor of a pose estimation model might return more than one instance (multi-person pose estimation). So I wondered if the function can be improved to accept more than one instance as a general case or make another function that works on more than one instance.

For my case, I tried testing this by making a function that accepts a Tensor of size (N, K, 3) where N is the number of instances, K is the number of keypoints instead of (K, 3), and loop over the instances. It worked well for me when I plotted the visualization. And I wonder if this can improve the Visualizer class by making another function specific to multiple instance pose estimation or editing the existing function.

Start with the Visualizer.draw_and_connect_keypoints entry point and compare its single-instance input handling with draw_instance_predictions. Determine whether the requested multi-instance pose visualization should extend the existing function or use a separate entry point, then verify that multiple keypoint instances render without unnecessary bounding boxes.

Written by the indexing model from the issue text.