MiDaS_RVC
Monocular Depth

Submitted by
Rene Ranftl
Authors
René Ranftl, Katrin Lasinger, David Hafner, Konrad Schindler, Vladlen Koltun
Venue
arXiv:1907.01341
Abstract
Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer. Standard pretrained model from Dec 2019 with resize method "minimal" for pre-processing (see code link) . Depths scaled and shifted to match the average scale and shift in the KITTI benchmark.
Runtime
100 ms
External data used
Default MiDaS model, not trained on VIPER
Links
Number of submissions
1

Detailed results
MetricAlldaysunsetsnowrainnight
Scale invariant logarithmic error 0.84 0.70 0.87 0.74 1.03 0.87
Absolute squared error [%] 17.99 14.86 13.87 10.34 25.84 25.03
Root mean squared error of the inverse depth 67.53 54.80 70.70 47.78 92.61 71.76