[Feature] Improve Visualizer Rendering Speed with OpenCV Backend

[Ben-Louis]

What is the feature?

The current rendering methods of Visualizer are slow, making it challenging to support real-time applications such as MMPose's webcam demonstrations. Could you please enhance the rendering methods by incorporating an OpenCV backend, similar to how the show method operates. We have tested the Visualizer rendering method, which includes drawing human poses on images using draw_circles and draw_lines functions, and found that the average speed for drawing a single human pose is 0.1 seconds. By utilizing OpenCV to implement the rendering methods, we have observed an increase in speed over 100 times faster.

Any other context?

No response

[Om-Doiphode]

Hi @Ben-Louis, I would like to work on this issue, can you guide me on how to get started with it?

[Ben-Louis]

Hi @Om-Doiphode, you might have to create a new class of Visualizer and overwrite the required methods using OpenCV. You can take reference from OpencvBackendVisualizer at open-mmlab/mmpose#2283 which has already implemented some methods. It might be helpful in guiding your implementation.