We adapted the diffusion-based depth estimation model Pixel-Perfect Depth to perform depth refinement.
You can run our trained model using the Colab notebook linked above.
source ./setup.sh # installs dependencies for repo and submodules
source ./script/data_fetch/data-fetch-small.sh # use data-fetch.sh for the full datasets
source ./script/data_fetch/construct_lists.sh # constructs data split txt files for train/val/test for each dataset.
All models are abstracted into a single function in ppd_sharpdepth/depth_estimators.py. Examples of running inference on a dataset can be found in infer.sh. Model outputs are dumped into the preds directory. Note that the model_architecture must match an enum string from the ModelArchitecture Enum in ppd_sharpdetph/depth_estimators.py
See eval.sh for examples for how to run inference for a given dataset.
We adapted the diffusion-based depth estimation model Pixel-Perfect Depth to perform depth refinement.
You can run our trained model using the Colab notebook linked above.
source ./setup.sh # installs dependencies for repo and submodules
source ./script/data_fetch/data-fetch-small.sh # use data-fetch.sh for the full datasets
source ./script/data_fetch/construct_lists.sh # constructs data split txt files for train/val/test for each dataset.
All models are abstracted into a single function in ppd_sharpdepth/depth_estimators.py. Examples of running inference on a dataset can be found in infer.sh. Model outputs are dumped into the preds directory. Note that the model_architecture must match an enum string from the ModelArchitecture Enum in ppd_sharpdetph/depth_estimators.py
See eval.sh for examples for how to run inference for a given dataset.