[논문 리뷰] From Pixels to Graphs: Open-Vocabulary Scene Graph Generation with Vision-Language Models

PreliminaryScene Graph GenerationScene Graph Generation의 목적은 visual relationships $\mathcal{R}=\{r_{ij}\}_{i\neq j}$와 $N^{v}$ entities $\mathcal{V}\{v_{i}\}^{N^{v}}_{i=1}$로 구성된 scene graph $\mathcal{G}_{sg}=\{\mathcal{V}_{e},\mathcal{R}\}$ 를 생성하는 것이다.relation triplet $r_{ij}=(\mathbf{v}_{i},c^{e}_{ij},\mathbf{v}_{j})$는 predicate category $c^{e}_{ij}$로 표기된 $i$번째와 $j$번째 entities 사이의 relationship을 ..

• format_list_bulleted Paper
• · 2025. 4. 3.
• textsms

[논문 리뷰] Fine-Grained Scene Graph Generation with Data Transfer

IntroFull MechanismInternal Data TransferConfusion pair discovery : specify confused predicate pairs as potential general-informative pairs for given subject and object classes.Transfer Pair Judgement : judge whether the candidate pair is validTriplet Transfer : transfer data from the selected general predicate class to the corresponding informative oneConfusion Pair DiscoveryRelation이 subject와 ..

• format_list_bulleted Paper
• · 2025. 4. 2.
• textsms