Add Functions for Visualizing COCO Predictions and Ground Truth

Summary

This PR introduces two new functions to simplify the visualization of COCO-format predictions and their comparison with ground truth (GT) annotations.

Changes

1. visualize_coco_predictions:

   • Visualizes predictions from a COCO-format dataset.
   • Inputs:
     • predictions: An object in the same format as returned by evaluate(*).
     • test_file_json: Path to the COCO-format test set JSON file with GT annotations.
     • num_samples: Number of samples to visualize.
     • output_dir: Directory to save visualization outputs.
     • draw_skeleton: Option to draw skeleton connections between keypoints.
   • Functionality:
     • Matches predictions to corresponding ground truth annotations.
     • Utilizes the plot_gt_and_predictions function for visualization.

2. plot_gt_and_predictions:

   • Handles detailed plotting of ground truth and predictions on a single image.
   • Inputs:
     • image_path: Path to the image
     • gt_bodyparts, pred_bodyparts: Keypoint arrays for ground truth and predictions.
     • output_dir: Directory to save the labeled images.
     • Additional customization options like dot_size, colormap, and confidence thresholds.
   • Functionality:
     • Supports various visualization modes (bodypart or individual coloring).
     • Plot the gt and predictions on the given image.

Motivation

• Improves debugging and performance analysis by offering a clear visual comparison between predictions and ground truth.
• Simplifies workflows for analyzing COCO-format datasets.

Testing

• The functions were tested on sample datasets and worked as expected.
  
  

Acknowledgments

Special thanks to @n-poulsen for his valuable guidance and support!