Skip to content

Latest commit

 

History

History
81 lines (57 loc) · 5.59 KB

README.md

File metadata and controls

81 lines (57 loc) · 5.59 KB

Computer Vision Annotation Custom Window

This custom window enables you to annotate the results of an object detection model or a pose (keypoint) detection model. This enables you to visualize the results of the model. The custom window annotates the bounding boxes of object detections and adds circles for the keypoints. For object detections, the custom window also displays the detected class and probability. If object tracking information is available, the object ID is also displayed. Each object ID is displayed in a different color.

Installation

Upload the annotation.py configuration file to SAS Event Stream Processing Studio. For more information, see Upload a Configuration File in SAS Help Center.

Example Output

Usage

Input Variables

Fields related to image and object detection are required. Fields related to keypoints are optional. Object ID (for object tracking) and attributes are also optional.

Name Description Type Required or Optional
image Input image blob Required
label Delimited list containing the class of the detected objects string Required
x Top left X-coordinates of the bounding boxes array(dbl) Required
y Top left Y-coordinates of the bounding boxes array(dbl) Required
w Widths of the bounding boxes array(dbl) Required
h Heights of the bounding boxes array(dbl) Required
score Confidence scores array array(dbl) Required
object_id Tracked object ID, for example from the Object Tracker window array(i32) Optional
attribute Delimited list containing the attribute(s) of the detected objects string Optional
object_track_count Number of tracks per detected object array(i32) Optional
object_track_kpts_count Number of keypoints per detected object for the track array(i32) Optional
object_track_kpts_x X-coordinates for the keypoints track array(dbl) Optional
object_track_kpts_y Y-coordinates for the keypoints track array(dbl) Optional
object_track_kpts_score Confidence scores for the keypoints track array(dbl) Optional
object_track_kpts_label_id Label IDs for the keypoints track array(i32) Optional

Output Variables

Add an output field, of type blob, to store the annotated image. If you use the same image as used in the input variables, the original image is overwritten with an annotated image.

Name Description Type
annotated_image Annotated image blob

Initialization

Set the options for the custom window. Note that either png or jpg is needed as the value for output_image_encoding to display images in Grafana.

Name Description Default
input_image_encoding Input image encoding - must be one of the following: wide, jpg, png wide
output_image_encoding Output image encoding - must be one of the following: wide, jpg, png jpg
object_label_separator Object label separator ,
kpts_labels Keypoint labels ``

An example value for kpts_labels is nose,l_eye,r_eye,l_ear,r_ear,l_shoulder,r_shoulder,l_elbow,r_elbow,l_wrist,r_wrist,l_hip,r_hip,l_knee,r_knee,l_ankle,r_ankle for YOLOv7 Pose.

Development

Real-life Test

This custom window has been tested using the Pose Estimation Using an ONNX Model (YOLO Pose Version 7) example.

Unit Tests

In the root directory, run

python -m unittest discover -v -b

The test_output/ folder will contain the output of the tests.

Docstrings

The docstrings are formatted with pydocstringformatter.

  • Install: pip install --upgrade pydocstringformatter
  • Usage: pydocstringformatter -w --no-final-period --no-capitalize-first-letter *.py

Future Ideas

  • Write outcomes to JPG or MP4 files, by using a subscriber connector