v2.0.0 - Rosbag2 Anonymizer

[2.0.0] - 2024-11-07

Release Notes

Key Changes:

• add validation code
• add validation documentation
• add time and requirements documentation

Other Changes:

• fix time consumption in blur function
• fix time consumption Grounding DINO pre-process
• fix empty qos profile issue

v1.0.0 - Rosbag2 Anonymizer

[1.0.0] - 2024-10-16

Release Notes

Key Changes:

• Removed Models:

  • Yolov8x
  • Segment Anything (SAM)

• Added Models:

  • Yolo11x
  • Segment Anything 2 (SAM-2)

Performance and Inference Time Comparisons of Updated Models

• Since there is not to much performance difference between segment-anything-2-large and segment-anything-2-small, we used segment-anything-2-small to reduce inference time. Here is the inference times of different segment anything models on single image:

  • segment-anything-large: ~480 ms (old) (model size: 2.6 gb)
  • segment-anything-2-large: ~210 ms (model size: 898.1 mb)
  • segment-anything-2-small: ~110 ms (newly added) (model size: 184.4 mb)

• Newly added yolo11x model trained with dataset which also used for yolov8x in past. There is no big performance differences. (Videos only contains yolo outputs not UnifiedModel outputs.)

  • yolov8x: https://youtu.be/GeGrZvYU5GA (inference time ~16 ms per image)
  • yolo11x: https://youtu.be/IgHR5HhDNtA (inference time ~20 ms per image)

v0.0.0 - Rosbag2 Anonymizer

First release of Rosbag2 Anonymizer.

• This version includes first version of Rosbag2 Anonymizer tool and pre-trained YOLOv8 model which trained on dataset which includes faces and license plates