Unsupervised Keypoints from Pretrained Diffusion Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hedlin, Eric, Sharma, Gopal, Mahajan, Shweta, He, Xingzhe, Isack, Hossam, Rhodin, Abhishek Kar Helge, Tagliasacchi, Andrea, Yi, Kwang Moo
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911883588534272
author Hedlin, Eric
Sharma, Gopal
Mahajan, Shweta
He, Xingzhe
Isack, Hossam
Rhodin, Abhishek Kar Helge
Tagliasacchi, Andrea
Yi, Kwang Moo
author_facet Hedlin, Eric
Sharma, Gopal
Mahajan, Shweta
He, Xingzhe
Isack, Hossam
Rhodin, Abhishek Kar Helge
Tagliasacchi, Andrea
Yi, Kwang Moo
contents Unsupervised learning of keypoints and landmarks has seen significant progress with the help of modern neural network architectures, but performance is yet to match the supervised counterpart, making their practicability questionable. We leverage the emergent knowledge within text-to-image diffusion models, towards more robust unsupervised keypoints. Our core idea is to find text embeddings that would cause the generative model to consistently attend to compact regions in images (i.e. keypoints). To do so, we simply optimize the text embedding such that the cross-attention maps within the denoising network are localized as Gaussians with small standard deviations. We validate our performance on multiple datasets: the CelebA, CUB-200-2011, Tai-Chi-HD, DeepFashion, and Human3.6m datasets. We achieve significantly improved accuracy, sometimes even outperforming supervised ones, particularly for data that is non-aligned and less curated. Our code is publicly available and can be found through our project page: https://ubc-vision.github.io/StableKeypoints/
format Preprint
id arxiv_https___arxiv_org_abs_2312_00065
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Unsupervised Keypoints from Pretrained Diffusion Models
Hedlin, Eric
Sharma, Gopal
Mahajan, Shweta
He, Xingzhe
Isack, Hossam
Rhodin, Abhishek Kar Helge
Tagliasacchi, Andrea
Yi, Kwang Moo
Computer Vision and Pattern Recognition
Unsupervised learning of keypoints and landmarks has seen significant progress with the help of modern neural network architectures, but performance is yet to match the supervised counterpart, making their practicability questionable. We leverage the emergent knowledge within text-to-image diffusion models, towards more robust unsupervised keypoints. Our core idea is to find text embeddings that would cause the generative model to consistently attend to compact regions in images (i.e. keypoints). To do so, we simply optimize the text embedding such that the cross-attention maps within the denoising network are localized as Gaussians with small standard deviations. We validate our performance on multiple datasets: the CelebA, CUB-200-2011, Tai-Chi-HD, DeepFashion, and Human3.6m datasets. We achieve significantly improved accuracy, sometimes even outperforming supervised ones, particularly for data that is non-aligned and less curated. Our code is publicly available and can be found through our project page: https://ubc-vision.github.io/StableKeypoints/
title Unsupervised Keypoints from Pretrained Diffusion Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2312.00065