When I use opencv to picture the labels I get this :Īs u can see, the dots dont resemble the bounding box coordinates but it looks kinda familiar. The last two steps are there to reshape it, as i get the (x,y,z) coordinates but the z coordinate is not needed. In this gig, Ill take an image of your 2D character and turn it into an accurate and detailed 3D model. If you have a picture of your favorite 2D character and want to turn it into a realistic looking 3D character, then youve come to the right place. Here is the code to get the 2d bounding box scene = īb_vertices = The idea is to use a for loop to easily generate 1000 pictures so i can train the model well. I generate an object in blender, position its centroid at (0,0,0) and then take a picture of it. Unfortunaly it doesnt match the ground truth data, probably Im doing something wrong in my code.įirst of all im trying to recreate this 6D Pose Estimation Algorithm:
Its going pretty good but I need to save the labels aka the 2d coordinates of the bounding box showed in the picture. I was trying to use some 6D Pose Estimation algorithm with a custom dataset.
0 kommentar(er)
