Find a file

Laurent FAINSIN bb93f6ab47 💄 black autoformatting		2023-04-24 11:45:30 +02:00
cuda	🗑️ replace deprecated cuda instructions	2023-04-24 11:39:02 +02:00
.editorconfig	🔧 add .editorconfig	2023-04-24 11:43:35 +02:00
.gitattributes	🙈 add .gitattribute	2023-04-24 11:42:40 +02:00
.gitignore	🙈 better .gitignore	2023-04-24 11:37:16 +02:00
__init__.py	first commit	2019-09-10 01:05:04 -07:00
emd.py	💄 black autoformatting	2023-04-24 11:45:30 +02:00
pyproject.toml	💄 black autoformatting	2023-04-24 11:45:30 +02:00
README.md	update	2019-09-10 01:06:14 -07:00
setup.py	💄 black autoformatting	2023-04-24 11:45:30 +02:00
test_emd_loss.py	💄 black autoformatting	2023-04-24 11:45:30 +02:00

PyTorch Wrapper for Point-cloud Earth-Mover-Distance (EMD)

Dependency

The code has been tested on Ubuntu 16.04, PyTorch 1.1.0, CUDA 9.0.

First compile using

    python setup.py install

Then, copy the lib file out to the main directory,

    cp build/lib.linux-x86_64-3.6/emd_cuda.cpython-36m-x86_64-linux-gnu.so .

Then, you can use it by simply

    from emd import earth_mover_distance
    d = earth_mover_distance(p1, p2, transpose=False)  # p1: B x N1 x 3, p2: B x N2 x 3

Check test_emd_loss.py for example.

The cuda code is originally written by Haoqiang Fan. The PyTorch wrapper is written by Kaichun Mo. Also, Jiayuan Gu provided helps.

MIT