Saved in:
Bibliographic Details
Main Authors: Chen, Zhengting, Cheng, Lei, Ding, Lianghui, Lin, Liang, Zhang, Quanshi
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2410.04421
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912610192982016
author Chen, Zhengting
Cheng, Lei
Ding, Lianghui
Lin, Liang
Zhang, Quanshi
author_facet Chen, Zhengting
Cheng, Lei
Ding, Lianghui
Lin, Liang
Zhang, Quanshi
contents This paper explains a neural network for image generation from a new perspective, i.e., explaining representation structures for image generation. We propose a set of desirable properties to define the representation structure of a neural network for image generation, including feature completeness, spatial boundedness and consistency. These properties enable us to propose a method for disentangling primitive feature components from the intermediate-layer features, where each feature component generates a primitive regional pattern covering multiple image patches. In this way, the generation of the entire image can be explained as a superposition of these feature components. We prove that these feature components, which satisfy the feature completeness property and the linear additivity property (derived from the feature completeness, spatial boundedness, and consistency properties), can be computed as OR Harsanyi interaction. Experiments have verified the faithfulness of the disentangled primitive regional patterns.
format Preprint
id arxiv_https___arxiv_org_abs_2410_04421
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Disentangling Regional Primitives for Image Generation
Chen, Zhengting
Cheng, Lei
Ding, Lianghui
Lin, Liang
Zhang, Quanshi
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
This paper explains a neural network for image generation from a new perspective, i.e., explaining representation structures for image generation. We propose a set of desirable properties to define the representation structure of a neural network for image generation, including feature completeness, spatial boundedness and consistency. These properties enable us to propose a method for disentangling primitive feature components from the intermediate-layer features, where each feature component generates a primitive regional pattern covering multiple image patches. In this way, the generation of the entire image can be explained as a superposition of these feature components. We prove that these feature components, which satisfy the feature completeness property and the linear additivity property (derived from the feature completeness, spatial boundedness, and consistency properties), can be computed as OR Harsanyi interaction. Experiments have verified the faithfulness of the disentangled primitive regional patterns.
title Disentangling Regional Primitives for Image Generation
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2410.04421