Skip to content

Data preparation

On first usage of a dataset the script will download and preprocess the datasets automatically. This is the recommended way to use the package as it ensures an unmodified dataset.

If you already downloaded a dataset and want to use symlinks instead of storing them again to save storage space, you can follow the manual instructions below.

The default directories can be found in the configuration file for the respective dataset (REAL275, REDWOOD75).

REAL275

For download links check the NOCS repository.

The expected directory structure for REAL275 evaluation is as follows: 

    {root_dir}/real_test/...
    {root_dir}/gts/...
    {root_dir}/obj_models/...
An additional directory {root_dir}/cpas_toolbox/ will be created to store preprocessed files. By default {root_dir} will be data/nocs/ (i.e., relative to the current working directory, when executing the evaluation script), but it can be modified.

REDWOOD75

To download the raw data check the redwood-data website. You can download the REDWOOD75 annotations here. Only the Redwood sequences ids contained in this file are required for evaluation.

The expected directory structure for REDWOOD75 evaluation is as follows: 

    {root_dir}/bottle/rgbd/00049/depth/*.png
    {root_dir}/bottle/rgbd/...
    {root_dir}/bowl/...
    {root_dir}/mug/...
    {ann_dir}/obj_models/...
By default {root_dir} will be data/redwood/ (i.e., relative to the current working directory, when executing the evaluation script) {ann_dir} will be data/redwood75/, but those can be modified.