search

cleinc / bts

From Big to Small: Multi-Scale Local Planar Guidance for Monocular Depth Estimation
GNU General Public License v3.0
626 stars 177 forks source link

Point cloud distance #69

Open surajmahangade opened 4 years ago

surajmahangade commented 4 years ago

Hello It is great work. I am using your model to calculate depth and from depth i am calculating point cloud. So when i move the camera forward i expect that the point cloud will also come closer to the camera,but it won't come close(to the camera).So is the depth image values created is only between a particular range. I am using your bts_nyu_v2_pytorch_densenet161 model. I don't know if i am clear. My question is can the values of the depth be used directly for calculating distance of a real object. Or it just calculates depth from a frame and is not related with actual distance. I am calculating distance using this with the pre-processing (all the calculations are right because the depth image looks good ,but just it is not giving me real depth values) lpg8x8, lpg4x4, lpg2x2, reduc1x1, depth_est = model(image_cropped, focal) Then i just scale it using pred_depth_scaled = depth*2500

If you need any other information let me know. Thank you.

mathmax12 commented 3 years ago

Hi I am doing the similar thing. What is the "focal" value you feed to the network?