OpenCV unsupported depth of input image, 'depth' is 3 (CV_16S) when using cv2.cvtColor-CodePudding

I'm reading in a video file and convert it to a numpy array.

like so:

video= np.array(skvideo.io.vread(folder   "video.avi"), dtype=np.int16)

later I access the first frame and save it under img

img = video[0]

the shape now has 3 channels:

389, 516, 3

When converting to grayscale:

gray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)

I get this error:

error: OpenCV(4.1.2) /io/opencv/modules/imgproc/src/color.simd_helpers.hpp:94: error: (-2:Unspecified error) in function 'cv::impl::{anonymous}::CvtHelper<VScn, VDcn, VDepth, sizePolicy>::CvtHelper(cv::InputArray, cv::OutputArray, int) [with VScn = cv::impl::{anonymous}::Set<3, 4>; VDcn = cv::impl::{anonymous}::Set<1>; VDepth = cv::impl::{anonymous}::Set<0, 2, 5>; cv::impl::{anonymous}::SizePolicy sizePolicy = (cv::impl::<unnamed>::SizePolicy)2u; cv::InputArray = const cv::_InputArray&; cv::OutputArray = const cv::_OutputArray&]'
> Unsupported depth of input image:
>     'VDepth::contains(depth)'
> where
>     'depth' is 3 (CV_16S)

The video and images display fine. I actually got to this error because I wanted to plot histograms and I get an error with no documentation when passing the first frame.

Is it possible that I have to use cv2.VideoCapture? I don't see how this should be the case as it's a perfectly fine numpy array.

thank you for your help!

CodePudding user response：

cvtColor complains because the data has the wrong dtype.

Your data has dtype int16 (signed integer).

cvtColor only accepts uint8, uint16, and float32. Those are CV_8U, CV_16U, and CV_32F in OpenCV parlance, which have the numerical values 0, 2, 5.

Solution: use a supported dtype. It seems that skvideo.io.vread already returns data in uint8. Just drop the superfluous np.array(..., dtype=np.int16) around it.

CodePudding user response：

Weirdly, it does work if I read in the video with video capture and then iterate over the frames.

So one solution to above is

video= cv2.VideoCapture(folder   'video.avi')
is_true, frame = video.read()

then do the rest as before.

If someone could explain why (what the difference between the arrays is) that'd be great.