Getting Started

Clone Repository

To begin with, clone the RoboDepth repository to the desired location:

cd models/  # desired location

git clone https://github.com/ldkong1205/RoboDepth.git

Robustness Probing

Our benchmark currently supports the following depth estimation algorithms:

• Supervised Depth Estimation

  • DORN
  • BTS
  • AdaBins
  • DPT
  • LapDepth
  • SimIPU
  • NeWCRFs
  • GLPDepth
  • DepthFormer

• Self-Supervised Depth Estimation

  • MonoDepth2
  • DepthHints
  • FeatDepth
  • PackNet-SfM
  • HR-Depth
  • ManyDepth
  • EPCDepth
  • DIFFNet
  • CADepth
  • MonoDEVSNet
  • DynamicDepth
  • Dyna-DM

• Semi-Supervised Depth Estimation

  • SemiDepth

MonoDepth2, ICCV'19

To evaluate the robustness of MonoDepth2 under common corruptions, run the following:

cd zoo/MonoDepth2

sh evaluate_kittic.sh
# --load_weights_folder: "path to the pretrained weights"
# --eval_mono or --eval_stereo: "set eval mode for mono and stereo"

DIFFNet, BMVC'21

To evaluate the robustness of DIFFNet under common corruptions, run the following:

cd zoo/DIFFNet

sh evaluate_kittic.sh
# --load_weights_folder: "path to the pretrained weights"
# --eval_mono or --eval_stereo: "set eval mode for mono and stereo"