Darknet: Detect numpy image with Python

[louisondumont]

The script darknet.py loads the image directly with a filepath. I'd like to use the detect function with a numpy image loaded from OpenCV. How can I do this? Thanks.

[saihv]

You can convert a numpy array to the format Darknet expects using this function (borrowed from this script you can use as reference)

def array_to_image(arr):
    import numpy as np
    # need to return old values to avoid python freeing memory
    arr = arr.transpose(2,0,1)
    c = arr.shape[0]
    h = arr.shape[1]
    w = arr.shape[2]
    arr = np.ascontiguousarray(arr.flat, dtype=np.float32) / 255.0
    data = arr.ctypes.data_as(POINTER(c_float))
    im = IMAGE(w,h,c,data)
    return im, arr

[louisondumont]

ctypes.ArgumentError: argument 2: <type 'exceptions.TypeError'>: expected IMAGE instance instead of tuple

the snippet you shared is coming from the original repo, hence it seems like the classes used aren't compatible with Alexey's repo.

[AlexeyAB]

@louisondumont I added it.

1. Just uncomment these lines: https://github.com/AlexeyAB/darknet/blob/573d7e80814a4cc3c08897f6c0f67ea189339856/darknet.py#L221-L223

2. And comment this line: https://github.com/AlexeyAB/darknet/blob/573d7e80814a4cc3c08897f6c0f67ea189339856/darknet.py#L260