MonoDepth-PyTorch

Why we included this project

This PyTorch port of MonoDepth reimplements the unsupervised monocular depth estimation method from Godard et al. You train a depth model from stereo image pairs without ground-truth depth labels. The encoder is pluggable, so you can swap in ResNet18, ResNet50, or any torchvision ResNet variant, and batch normalization was added to keep training stable. A pretrained ResNet18 model ships with the repo, along with a notebook that walks through training and testing on KITTI. The code is small enough to read end to end, which makes it a useful reference for reproducing the method or adapting it to your own camera setup.