Cuda HOG computed , conversion To image

Maths_Electronics_Tu · October 15, 2023, 4:21pm

Hi, now that Cuda is working, I am facing another challenge. i compute HOG features by Cuda, but the library does not convert the features to jpg image as CPU version does , can you support me to chaive this !

berak · October 15, 2023, 4:28pm

what do you mean ? (sounds you misunderstood something)
what would you need jpg images of hog features for ?

please explain…

Maths_Electronics_Tu · October 15, 2023, 4:37pm

the cpu hog OpenCV can return hog features and images that visualize the features , but Cuda returns only the features , so I need to code and to save the features as gray image .

berak · October 15, 2023, 4:59pm

what is your use-case for HOG ?
i still doubt, that you need an image visualization of the features, but if so, download to cpu & process as usual

Maths_Electronics_Tu · October 15, 2023, 5:10pm

I use it to train multimodal models; it takes a long time by the CPU but is quick with Cuda; Cuda returns a 2D array, but the documentation does not provide a detailed explanation.

berak · October 15, 2023, 6:13pm

something like this for visualization ?

github.com

danielshaving/CV_Features_HoG_Feature_Extraction/blob/11a299038e5ed551e5c74fc2c668f3387aabac2c/Hog Feature by pure numpy.py#L74-L91


      
          def render_gradient(self, image, cell_gradient):
              cell_width = self.cell_size / 2
              max_mag = np.array(cell_gradient).max()
              for x in range(cell_gradient.shape[0]):
                  for y in range(cell_gradient.shape[1]):
                      cell_grad = cell_gradient[x][y]
                      cell_grad /= max_mag
                      angle = 0
                      angle_gap = self.angle_unit
                      for magnitude in cell_grad:
                          angle_radian = math.radians(angle)
                          x1 = int(x * self.cell_size + magnitude * cell_width * math.cos(angle_radian))
                          y1 = int(y * self.cell_size + magnitude * cell_width * math.sin(angle_radian))
                          x2 = int(x * self.cell_size - magnitude * cell_width * math.cos(angle_radian))
                          y2 = int(y * self.cell_size - magnitude * cell_width * math.sin(angle_radian))
                          cv2.line(image, (y1, x1), (y2, x2), int(255 * math.sqrt(magnitude)))
                          angle += angle_gap
              return image

i’d probably use the (1d) vector of histograms for that, not a drawn visualization image (far too much redundancy)

Maths_Electronics_Tu · October 15, 2023, 6:17pm

i need this format to use the same backbone for rgb image and optical flow so i just reduced the RGB 3 chs to one

Topic		Replies	Views
Question in HOGDescriptor::computeGradient soure code C++ objdetect	1	379	July 5, 2022
Cuda operations give black frames C++ cuda , imgproc	7	786	March 21, 2022
Imshow() and imwrite() outputs are different and normalization is not preserving original image Python highgui , imgcodecs	2	1877	April 22, 2021
Hi…I am very much eager to convert the following code into opencv GPU, what is the correct way to do that? Python	4	203	November 17, 2021
Normalizing CF_32 images with negative pixel values gives output with some black regions of the original image turning white Python highgui , cuda , imgcodecs	5	1931	April 23, 2021

Cuda HOG computed , conversion To image

Related Topics