shared-scripts

Commonly used scripts to share.

Visualization

visualize.py can be used to quickly visualize mesh, point cloud, voxel, part box assembly and save rendered image if specified. Both one object or a directory or objects is supported.

python visualize.py --format {file format} --src {object path or directory to visualize} --save_image[optional]

Mesh is visualized by trimesh, point cloud by open3d and voxel/box by matplotlib.

Loss and Evaluation

loss-and-evaluation provides code for some distance metrics.

IoU

• iou.py provides functions to calculate IoU for binary arrays or axis-aligned bounding boxes. Implemented in numpy but should be easy converted to pytorch or tensorflow.

Chamfer Distance

• chamfer.py provides functions to calculate symmetric chamfer distance, implemented in numpy.
• https://github.com/chrdiller/pyTorchChamferDistance provides fast implementation, wrapped in pytorch.

EMD Distance

• https://github.com/daerduoCarey/PyTorchEMD provides fast implementationm, wrapped in pytorch. (But cannot parralleled over multiple GPUs..)
• https://github.com/optas/latent_3d_points/tree/master/external/structural_losses has fast tensorflow implementation. Chamfer distance is also included.

Hausdorff Distance

• directed_hausdorff.py provides pytorch implementation for directed hausdorff distance.
• Tensorflow implementation can be found here.

Light Field Distance

• pending...