Standalone Evaluation Pipeline

Now this sub-directory is not fully standalone, which has dependencies under the project root, should clean the dependency there later.

1. Inference trained model to generate objects: gen_diffusion.ipynb. This notebook assume you have a trained model saved under log dir. After running this notebook, you will see two folders under log/test/: G saved all generated objects. Viz have some gifs visualization. You should also have to run save_gt.ipynb to save the ground truth objects for evaluation.

2. Sample point clouds from generated object with sample_pcl.py. Here sample_pcl.sh is an example. After running this, you will see sampled point cloud npz files under PCL dir

3. Compute the instantiation distance (very slow) based on the saved pcl files with instantiation_distance.py and example in compute_id.sh. After several hours GPU computing, you will see the distance matrices saved underID_D_matrix.

4. Finally, you can compute the metrics from the saved distance matrices with compute_metrics.ipynb.