$D^2$ EFS Deep Depth Estimation For Stereo Cameras

Deep Learning model for stereo depth estimation. The model employs UNet architecture for fine depth estimation. The model allows you to either use the transpose convolution to reconstruct the depth image or use bilinear interpolation with same-padded CNNs. The pretrained model uses the version with the transpose convolution as it was the one yieled the highest accuracy.

How to use

1. Run python -m venv venv in the project directory
2. On Linux use source venv/bin/activate. on Windows, use venv\Scripts\activate To use the notebooks, make sure that you have anaconda installed and only run the first command in the previous list after being in the base environment.

Model Parameters

Download the model pretrained parameters from here