CenterPose Eval for Objectron

Overview

For a fair comparison with other methods on the Objectron benchmark, we develop our evaluation code based on the officially released evaluation code.

We offer two different variants:

• eval_image_official.py runs on the original officially released preprocessed dataset. (Not used in our experiment.)

• eval_video_official.py runs on the re-sorted officially released preprocessed dataset.

Note that all the load_model paths have to be re-configured to your new location.

For more evaluation options, please refer to eval_opts.py

Before you start evaluation process (only for eval_video_official.py), please prepare the dataset first, where we provide two scripts:

• download_test_video.py downloads the officially released preprocessed dataset.

• prepare_test_video.py re-sorts the shuffled officially released preprocessed dataset to arrange them into videos.

Utils

group_report_new.py collects the results from videos. It also provides the option to ignore specific samples or only collect the result from specific samples.

Evaluation with pre-defined configurations

To evaluate on multiple categories, we wrap the evaluation code into two scripts:

shell_eval_image_CenterPose.py runs on the original officially released preprocessed dataset (image). (We do not use it for our paper.)

shell_eval_video_CenterPose.py and shell_eval_video_CenterPoseTrack.py run on the re-sorted officially released preprocessed dataset (video).