RA-SGG: Retrieval-Augmented Scene Graph Generation Framework via Multi-Prototype Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yoon, Kanghoon, Kim, Kibum, Jeon, Jaehyung, In, Yeonjun, Kim, Donghyun, Park, Chanyoung
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915068948512768
author Yoon, Kanghoon
Kim, Kibum
Jeon, Jaehyung
In, Yeonjun
Kim, Donghyun
Park, Chanyoung
author_facet Yoon, Kanghoon
Kim, Kibum
Jeon, Jaehyung
In, Yeonjun
Kim, Donghyun
Park, Chanyoung
contents Scene Graph Generation (SGG) research has suffered from two fundamental challenges: the long-tailed predicate distribution and semantic ambiguity between predicates. These challenges lead to a bias towards head predicates in SGG models, favoring dominant general predicates while overlooking fine-grained predicates. In this paper, we address the challenges of SGG by framing it as multi-label classification problem with partial annotation, where relevant labels of fine-grained predicates are missing. Under the new frame, we propose Retrieval-Augmented Scene Graph Generation (RA-SGG), which identifies potential instances to be multi-labeled and enriches the single-label with multi-labels that are semantically similar to the original label by retrieving relevant samples from our established memory bank. Based on augmented relations (i.e., discovered multi-labels), we apply multi-prototype learning to train our SGG model. Several comprehensive experiments have demonstrated that RA-SGG outperforms state-of-the-art baselines by up to 3.6% on VG and 5.9% on GQA, particularly in terms of F@K, showing that RA-SGG effectively alleviates the issue of biased prediction caused by the long-tailed distribution and semantic ambiguity of predicates.
format Preprint
id arxiv_https___arxiv_org_abs_2412_12788
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle RA-SGG: Retrieval-Augmented Scene Graph Generation Framework via Multi-Prototype Learning
Yoon, Kanghoon
Kim, Kibum
Jeon, Jaehyung
In, Yeonjun
Kim, Donghyun
Park, Chanyoung
Computer Vision and Pattern Recognition
Scene Graph Generation (SGG) research has suffered from two fundamental challenges: the long-tailed predicate distribution and semantic ambiguity between predicates. These challenges lead to a bias towards head predicates in SGG models, favoring dominant general predicates while overlooking fine-grained predicates. In this paper, we address the challenges of SGG by framing it as multi-label classification problem with partial annotation, where relevant labels of fine-grained predicates are missing. Under the new frame, we propose Retrieval-Augmented Scene Graph Generation (RA-SGG), which identifies potential instances to be multi-labeled and enriches the single-label with multi-labels that are semantically similar to the original label by retrieving relevant samples from our established memory bank. Based on augmented relations (i.e., discovered multi-labels), we apply multi-prototype learning to train our SGG model. Several comprehensive experiments have demonstrated that RA-SGG outperforms state-of-the-art baselines by up to 3.6% on VG and 5.9% on GQA, particularly in terms of F@K, showing that RA-SGG effectively alleviates the issue of biased prediction caused by the long-tailed distribution and semantic ambiguity of predicates.
title RA-SGG: Retrieval-Augmented Scene Graph Generation Framework via Multi-Prototype Learning
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2412.12788