aisingapore / PeekingDuck

A modular framework built to simplify Computer Vision inference workloads.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Incorrect description for `count` input/output type

leeping-ng opened this issue · comments

count which is currently an output from dabble.bbox_count should just be an int type, but the description in RTD is incorrect:

count (List[int]): A list of integers representing the count of a pre-selected object (for example, ‘people’) detected in each frame through bboxes.

It could have been the initial intention to present it as a list to show different counts for different classes, but with the update to draw.legend, it was changed to count across all bboxes for simplicity. Going forward with new nodes coming out in v1.2 which will make use of count, suggest retaining it as an int type and updating the docs.

Here, it shows count is a single int of 3 even with 2 classes, "dog" and "person".

'bboxes': array([[0.31895882, 0.6725785 , 0.5876959 , 0.97450596],
        [0.44654298, 0.07115207, 0.7607977 , 0.7644112 ],
        [0.22748163, 0.04981984, 0.50799876, 0.6592544 ]], dtype=float32),
 'bbox_labels': array(['dog', 'person', 'person'], dtype='<U6'),
 'bbox_scores': array([0.72365254, 0.63344693, 0.39939374], dtype=float32),
 'count': 3}

This has been updated in the pull request for the CSRnet node. Once it is merged into dev, this issue will be closed.