Final project for CS 7643 Deep Learning focusing on reproducing the results detailed in "Self-Supervised Monocular Scene Decomposition and Depth Estimation" - S. Safadoust, F. Guney (2021)
torchvision-0.11.1pytorch-1.10.0numpy-1.21.2
Command for Training:
python train.py --png --data_path ../kitti_data --load_weights_folder /$path_to_weights$
Command for Evaluation:
python depth_masking_validation.py --png --data_path /$path_to_kitti_data$ --load_weights_folder /$path_to_weights$ --num_val_batches=2
Command for Viewing Logs:
tensorboard --logdir=<dir_of_log>
This codebase reproduces the MonoDepthSeg architecture described in the following work:
Self-Supervised Monocular Scene Decomposition and Depth Estimation
No public code implementations for this work currently exist.
This codebase also uses model architectures and utility functions from the following work per their non-commercial use license:
Digging into Self-Supervised Monocular Depth Prediction
Clément Godard, Oisin Mac Aodha, Michael Firman and Gabriel J. Brostow
nianticlabs/monodepth2 (Github)
Copyright © Niantic, Inc. 2018. Patent Pending.