Der Luftmensch's profile - activity

berak is right. Use cv::transform, you misunderstand the documentation.

If you have your distortion maps, remap doesn't care what you give it. So pass an iota image (rows or cols) for the dime

Look at using FFT/DFTs. They transform convolution to element-wise multiplication, offering a huge speedup over spatial

Look at median flow paper. There is apparently no benefit to using GFFT or FAST. You must also use more than one feature

The number of pyramid levels may be insufficient for the distances required going from image 1 to image 3. You might als

You number of pyramid levels may be insufficient for the distances required going from image 1 to image 3. You might als

You could do blending over ROIs, maintaining a total light image. So rather than call cv::circle() directly on the image

Do you need a multi-scale detector? Are your images at the same scale? If so, just use FAST for detection and FREAK for

Orb does multi-scale FAST, are you sure the points in close proximity aren't at different scales? Is non-maxima suppress

Why do you put & in front of channel argument? There is usually no reason to pass cv::Mat pointers.

Do not call release(), let it go out of scope.

crash in Mat::release() Hi, I have a function that creates a histogram: double median_(const cv::Mat& channel) {

Have you tried using GDAL directly rather than OpenCV? It will offer you all possible features compared to what little i

A Generalization of Otsu's Method and Minimum Error Thresholding. See the end of the paper for Python code.

If the camera is (0,0,0) of your coordinate frame, use cv::reprojectImageTo3D(). This requires the Q matrix from calibra

Try something similar to the following: // // This is a deep copy, OpenCV has no CV_64(U|S) types, only CV_32S: // cv::

As I said above, there is no int64 support in cv::Mat. Have you measured the performance costs of casting to int32? It m

Please elaborate on 'world coordinates'.

You need to scale the results, so multiply or divide by 255 or 65536, etc. Look more closely at that function.

It that case it looks right to me except that you might have rows and cols swapped. Are you seeing a problem? If so, can

If you can capture from the camera at 16-bits (even if dynamic range is 12-bits), that might be the simplest and best op

If your raw data is 12-bit pixels, interpreting them as 16-bits is not going to work and may seg fault. You need to firs

At present there is no CV_64S support, so your best option is to convert/cast to CV_32S for down stream processing.

Everyone will answer "deep learning", Haar cascades will almost certainly not work for anything close to what you say yo

Your image is 8-bits, 0-255, so normalizing with minmax 0-255 will likely do nothing. 0 will still be 0 and 255 will sti

No, this is how to make best use of the OpenCV API to get what I want with simplicity.

The image is 8-bits so values should already be 0-255, right? The operation won't change pixel values.

A few erodes followed by some dilates will coarsen the contour.

@berak, yes, this could be a 1D loop assuming continuous. The reason is to do this is for semantic segmentation. I have

@berak, yes, this could be a 1D loop assuming continuous. The reason is to do this is for semantic segmentation. I have

One Hot Encode function I have a matrix where each element is a class ID. I need to transform this matrix into a one-hot

Another option is tracking by detection, where you run your insect detector every X frames and associate new detections

Is this really an OpenCV issue? What do you mean by system buff/cache? Memory? Can you post a minimal example of your co

I was just wondering why resize if not for the disparity computation, but maybe they are very large and want to see them

They are also vertically misaligned. Recalibrate and visualize the undistorted and rectified images as an anaglyph to ve

Your left image is out of focus. You don't used resized images in the SGBM computation? Your SGBM parameters are a littl

Take a look at the source code. For instance, copy and modify the function ICAngles to suit your needs.

DSI = Disparity Space Image (HxWxD). This is the volume you aggregate over for a smoothed cost estimate. This is also th

If you have a DSI, every comparison you need to do R-L you already did for L-R.

If you have a DSI, every comparison you need to R-L you already did for L-R.

The following has worked well for me, with no unnecessary data allocation: void hwc_to_chw(cv::InputArray src, cv::Outp

void hwc_to_chw(cv::InputArray src, cv::OutputArray dst) { const int src_h = src.rows(); const int src_w = s

void hwc_to_chw(cv::InputArray src, cv::OutputArray dst) { const int src_h = src.rows(); const int src_w =

void hwc_to_chw(cv::InputArray src, cv::OutputArray dst) { const int src_h = src.rows(); const int src_w =

Is there an OpenCV function to take a single channel image and extend it to N-channels, copying/repeating the pixel value to all channels, something like a pixel-wise broadcast?

I need to be able to per-element multiply a single channel image with an n-channel image and the only way to do this with high-level function calls is to extend/repeat the single channel image to match the dimensions of my n-channel image. Can I achieve this with a cv::Mat from cv::Mat::reshape() passed to cv::repeat()? Can I achieve this with cv::transform()?

I solved this using cv::transform(), see the example below to extend the single_channel image to n channels. cv::Ma

void hwc_to_chw(cv::InputArray src, cv::OutputArray dst) { const int src_h = src.rows();