1. In predictions/visualize_kitti.py, specify the kitti raw data directory via data_dir; specify the project root directory via project_dir.
2. For Godard, set mode to invdepth. For GeoNet, set mode to depth.
3. In predictions directory, interpolate the ground truth sparse depth by invoking

python visualize_kitti.py interp

4. Once you have the interpolated ground truth, run the visualization script as follows:

python visualize_kitti.py \
path/to/your/prediction/of/model/A.npy tag_for_model_A \
path/to/your/prediction/of/model/B.npy tag_for_model_B

Note, you need to pair the path to the npy file of the prediction and a tag for that model. For instance

python visualize_kitti.py \
predictions/GeoNet_depth.npy GeoNet \
predictions/GeoNet_SIGL_depth.npy GeoNet_SIGL