Frame Averaging
Project description
Frame Averaging - Pytorch (wip)
Pytorch implementation of a simple way to enable (Stochastic) Frame Averaging for any network. This technique was recently adopted by Prescient Design in AbDiffuser
Install
$ pip install frame-averaging-pytorch
Usage
import torch
from frame_averaging_pytorch import FrameAverage
# contrived neural network
net = torch.nn.Linear(3, 3)
# wrap the network with FrameAverage
net = FrameAverage(
net,
dim = 3, # defaults to 3 for spatial, but can be any value
stochastic = True # whether to use stochastic variant from FAENet (one frame sampled at random)
)
# pass your input to the network as usual
points = torch.randn(4, 1024, 3)
mask = torch.ones(4, 1024).bool()
out = net(points, frame_average_mask = mask)
out.shape # (4, 1024, 3)
# frame averaging is automatically taken care of, as though the network were unwrapped
or you can also carry it out manually
import torch
from frame_averaging_pytorch import FrameAverage
# contrived neural network
net = torch.nn.Linear(3, 3)
# frame average module without passing in network
fa = FrameAverage()
# pass the 3d points and mask to FrameAverage forward
points = torch.randn(4, 1024, 3)
mask = torch.ones(4, 1024).bool()
framed_inputs, frame_average_fn = fa(points, frame_average_mask = mask)
# network forward
net_out = net(framed_inputs)
# frame average
frame_averaged = frame_average_fn(net_out)
frame_averaged.shape # (4, 1024, 3)
Citations
@article{Puny2021FrameAF,
title = {Frame Averaging for Invariant and Equivariant Network Design},
author = {Omri Puny and Matan Atzmon and Heli Ben-Hamu and Edward James Smith and Ishan Misra and Aditya Grover and Yaron Lipman},
journal = {ArXiv},
year = {2021},
volume = {abs/2110.03336},
url = {https://api.semanticscholar.org/CorpusID:238419638}
}
@article{Duval2023FAENetFA,
title = {FAENet: Frame Averaging Equivariant GNN for Materials Modeling},
author = {Alexandre Duval and Victor Schmidt and Alex Hernandez Garcia and Santiago Miret and Fragkiskos D. Malliaros and Yoshua Bengio and David Rolnick},
journal = {ArXiv},
year = {2023},
volume = {abs/2305.05577},
url = {https://api.semanticscholar.org/CorpusID:258564608}
}
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
frame_averaging_pytorch-0.0.18.tar.gz
(220.7 kB
view hashes)
Built Distribution
Close
Hashes for frame_averaging_pytorch-0.0.18.tar.gz
Algorithm | Hash digest | |
---|---|---|
SHA256 | 6772c684bc99e3493ed085984f1080382999aceb7211166c6adaef7ae5c8c1c4 |
|
MD5 | 3cbf348f3fd34deefc33245dcf7a8717 |
|
BLAKE2b-256 | 2a41f164371e0302b764af392b000f1e506d1bfbdbec01e740539e73ea03b8bb |
Close
Hashes for frame_averaging_pytorch-0.0.18-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | eb1a8f79a4964710f7724cbcf6f3ea828fc270a50be911c78a4b7c7c0f870b53 |
|
MD5 | e3c7eb785c2d6152cdb062f43fc73796 |
|
BLAKE2b-256 | 5e04c4a51448d417bb2180087b1dcbe34057d86ea6b7454dfd61e4e11bd67564 |