Submitted by
Oliver Zendel
Authors
Vitor Guizilini and Rares Ambrus and Sudeep Pillai and Allan Raventos and Adrien Gaidon
Venue
CVPR 2020
Abstract
PackNet-SfM with PackNet01_HR_velsup_CStoK reference model; self-supervised training on KITTI and Cityscapes sequences; all inputs scaled to 1280x384; submitted as an RVC reference
Runtime
500 ms
External data used
KITTI Sequences; pretrained on Cityscapes
Links
Number of submissions
1

Detailed results
MetricAlldaysunsetsnowrainnight
Scale invariant logarithmic error 1.46 0.97 1.55 1.87 1.39 1.54
Absolute squared error [%] 50.25 23.66 51.82 64.40 44.09 67.27
Root mean squared error of the inverse depth 201.94 98.51 226.51 296.03 167.88 220.78