Name
Abstract | ||
|---|---|---|
It is always well believed that modeling the relative depth information between objects would be helpful for recognizing the spatial relations between pairs of objects in images, especially for the spatial relation like "behind" and "in front of." Nevertheless, there has not been evidence in support of the idea on spatial relation recognition. In this paper, we present a novel Depth-guided Spatial Relation Recognizer (DSRR) to predict the spatial predicate from object pairs under the umbrella of relative depth information in between. Particularly, DSRR capitalizes on the off-the-shelf depth estimator to predict the depth information for each object. The depth cues for each pair of objects are further integrated with language (object name) and 2D (bounding box coordinates) cues to perform spatial relation reasoning. Extensive experiments conducted on SpatialSense dataset validate our proposal and superior results are reported when comparing to state-of-the-art models. |
Year | DOI | Venue |
|---|---|---|
2020 | 10.1109/MIPR49039.2020.00065 | 2020 IEEE Conference on Multimedia Information Processing and Retrieval (MIPR) |
Keywords | DocType | ISBN |
spatial relation recognition,deep learning,depth estimation | Conference | 978-1-7281-4273-9 |
Citations | PageRank | References |
0 | 0.34 | 17 |
Authors | ||
5 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Xuewei Ding | 1 | 0 | 0.34 |
| Yehao Li | 2 | 75 | 8.57 |
| Yingwei Pan | 3 | 357 | 23.66 |
| Dan Zeng | 4 | 25 | 11.26 |
| Ting Yao | 5 | 842 | 52.62 |