Object Daddy

A simple, python-based object recognizer, basically an implementation of yolo-v3-tiny. This package is used for personal projects; I don't maintain it, so use at your own risk, and feel free to fork.

Usage

Basic usage is as follows. This project is built to work with rtsparty (also on pypi).

You can get and recognize one frame using the following:

from rtsparty import Stream
from objectdaddy import Daddy


stream = Stream()
daddy = Daddy()

frame = stream.get_frame()
results, frame = daddy.process_frame(frame)
for detection in results:
    detection.identify()

For continuous recognition, use the following code.

from rtsparty import Stream
from objectdaddy import Daddy


stream = Stream()
daddy = Daddy()


try:
    while True:
        frame = stream.get_frame()
        if stream.is_frame_empty(frame):
            continue
        daddy.process_frame(frame)
        for detection in daddy.get_current_detections():
            if not detection.has_been_processed_downstream:
                detection.identify()
                detection.has_been_processed_downstream = True
except KeyboardInterrupt:
    pass

The code makes the attempt to recognize objects across multiple frames, and keep a list of objects in memory accessed with the get_current_detections() function.

Notes

• This package supports only CPU-based inference.