Open Access
Table 2
A summary of scene graph construction
| Paper | Year | Category | Contribution |
|---|---|---|---|
| Yang et al[25] | 2018 | Object triplet model | First introduce the object triplet model to define the scene graph. Construct scene graph offline using the Visual Genome dataset |
| Nguyen et al[41] | 2019 | Object triplet model | Propose ReVec to reduce the complexity of GCN |
| Du et al[27] | 2020 | Object triplet model | Incorporate Faster R-CNN into the construction of the scene graph |
| Hu et al[32] | 2021 | Object triplet model | Introduce the Agent-Object KG representing the depth relationship between the agent and objects |
| Zhou et al[42] | 2021 | Object triplet model | Utilize the YOLOv3 for object detection, defining the spatial relationships between objects based on their relative directions |
| Qiu et al[43] | 2021 | Object triplet model | Propose a hierarchical object relationship learning approach by utilizing novel context vector as node embedding |
| Dang et al[31] | 2022 | Object triplet model | Introduce the DOA graph, employing attention mechanisms to enhance adaptability across unfamiliar environments |
| Zhou et al[44] | 2023 | Object triplet model | Introduce the heterogeneous relation graph constructed on multisource information fusion in the AI2-THOR environment |
| Lyu et al[45] | 2023 | Object triplet model | Combining global and local scene graph to capture global spatial information and local semantic relationship |
| Dang et al[46] | 2023 | Object triplet model | Propose a Target-Oriented Memory Graph (TOMG) to store the simplified agent state target orientation information |
| Yang et al[47] | 2024 | Object triplet model | Propose a time-varying KG network，coupling of scene graph, context vectors, and global spatial semantic information |
| Luo et al[48] | 2024 | Object triplet model | Construct indoor navigation Adaptive Relation Graph (ARG) without relying on any additional geometric information |
| Zhang et al[34] | 2021 | Zone triplet model | Introduce a HOZ graph, making spatial constraints like rooms and scenes |
| Singh et al[49] | 2023 | Zone triplet model | Propose the Scene Graph Contrastive (SGC) loss, which uses scene graph as training only supervisory signals |
| Xu et al[50] | 2024 | Zone triplet model | Leveraging visual-language pre-training to align natural language description with visual perception |
| Li et al[33] | 2021 | Instance triplet model | Tackle instance-level navigation tasks by developing an instance-level scene graph, with nodes incorporate color, texture and location |
| Kwon et al[39] | 2021 | Image triplet model | Introduce the VGM, directly utilizing the observation images for constructing the scene graph |
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.