alecda's profile - activity

opencv world not in build folder I have successfully built opencv version 4.4.0 with Cuda support Cuda toolkit v11.2 and

Error on forward method of cv2.dnn I am able to load the load a frozen graph but when I try to use the forward method I

Error on forward method of cv2.dnn I am able to load the load a frozen graph but when I try to use the forward method I

@berak will do, It is a lambda layer just performing image centering so img/255-.5

Error from cv2.dnn.readNetFromTensorflow Lambda layer I am encountering the following error when I try to make the forwa

I am attempting to calculate a surface normal map from my depth map using the following code:

import numpy as np
import cv2
depth=cv.imread(depthpath)
normals=np.zeros((depth.shape[0],depth.shape[1],3)
for x in range(0,depth.shape[0],1):
    for y in range(0,depth.shape[1],1):
        dzdx=(depth[x+1,y,0]-depth[x-1,y,0])/2
        dzdy=(depth[x,y+1,0]-depth[x,y-1,0])/2
        sub=np.asarray([-dzdx,-dzdy,1])
        normals[x,y,:]=sub/np.linalg.norm(sub)

this is based off the following thread: https://answers.opencv.org/question/8...

But I am getting a very odd output see the attached depth map and surface normal map:

Does anyone have any thoughts as to what is going on here?

sloppy mistakes fixed, cheers mate

@berak I would be accepting of a coarse approximation that looks decent like what can be seen in the linked post, I have

@berak obviously there are issues as we are not in the continuous setting. Furthermore, I would say there are even more

@berak Sure it is not anything different than the linked post: Reads in depth map initializes surface normal map loops o

Sure it is not anything different than the linked post: Reads in depth map initializes surface normal map loops over the

@berak even if I dont just do depth[x,y,0] and read in the way you suggest I get the same surface normal output...

calculating surface normal map from depth map weird errors: I am attempting to calculate a surface normal map from my de

sorry yes I know the code I actually used extracts a single channel since they are the same image stacked three times.

calculating surface normal map from depth map weird errors: I am attempting to calculate a surface normal map from my de

More than one input is Const Op System information (version) Detailed description I am trying to serve the following gi

OpenCV(4.1.2) Error: Assertion failed (nodesMapIt != nodesMap.end()) in cv::dnn::dnn4_v20190902::sortByExecutionOrder, I

OpenCV(4.1.2) Error: Assertion failed (nodesMapIt != nodesMap.end()) in cv::dnn::dnn4_v20190902::sortByExecutionOrder, I

I just wanted to make sure there was not any errors. I know this but sometimes c++ is weird about data types.

Can cv::dnn::Net take std::vector as argument? Say I have a pretrained model in tensorflow that accepts as input $(batch

errors regarding cv::dnn:readNetFromTensorflow I have a pretrained model in tensorflow and I converted the checkpoint da

errors regarding cv::dnn:readNetFromTensorflow I have a pretrained model in tensorflow and I converted the checkpoint da

opencv_world412d.dll was not found I have been trying to follow this guide to set up opencv with visual studio: https://