Closed antithing closed 4 years ago
The pre-trained models from Monodepth output metric depth (absolute distance in metre) after the scaling, i.e., depth = (focal_length baseline) / (image_width network_prediction)
Thank you! That's awesome. I have a fisheye stereo pair that i would like to use this one, would i need to train a new model?
The camera projection model for fisheye stereo pairs would be different from the pinhole projection model used in KITTI dataset. Therefore, you would need to train a new model.
Hi, and thank you for making this code available. This may be a stupid question, but will this code output real-world-scale distance values from the depth map?
Or is it up-to-scale, as it is with monocular slam?
Thanks!