[MLOps 1.5] Expand the built-ins: CV

[dbogunowicz]

Feature description

• Added built-in functions: detected_classes, number_detected_objects, mean_score_per_detection, std_score_per_detection.
• Removed a few built-ins, simplified a few, and added exhaustive unit tests.
• Now built-ins either return a single float/int or a dictionary (so that in the future we can easily extract the information from the built-in functions and send them to loggers via FunctionLogger).

Coverage:

As specified in the PRD:

cv_image_width
cv_image_height --> those two functions are covered by "image_shape"

cv_image_mean_pixel_r_channel
cv_image_mean_pixel_g_channel
cv_image_mean_pixel_b_channel --> those three functions are covered by "mean_pixels_per_channel"

cv_image_percent_0_pixels -> "fraction_zeros"

cv_num_classes_predicted - count of the number of unique classes in yolo/yolact -> "detected_classes"
cv_num_bboxes_predicted - count of the number of objects detected in yolo/yolact -> "number_detected_objects"

cv_mean_bbox_score - mean score for a prediction in yolo/yolact
cv_max_bbox_score - mean score for a prediction in yolo/yolact
cv_min_bbox_score - min score for a prediction in yolo/yolact -> covered by  "mean_score_per_detection" and "std_score_per_detection"

[corey-nm]

What does a configuration of actually using these look like? If I wanted to use a classes thing and an image thing I'd have to specify something like:

pipeline_outputs.image:
    - std_per_channel
pipeline_outputs.classes:
    - count_num_classes

?

[corey-nm]

lgtm, would be good to see some frontend examples to see how usable these all are

[dbogunowicz]

What does a configuration of actually using these look like? If I wanted to use a classes thing and an image thing I'd have to specify something like:

pipeline_outputs.image:
    - std_per_channel
pipeline_outputs.classes:
    - count_num_classes

?

pipeline_outputs.image:
    - func: std_per_channel
pipeline_outputs.classes:
    - func: count_num_classes

[bfineran]

looks great overall, few comments

[corey-nm]

lgtm!