Related Experiment Video
Updated: May 23, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
Cross-modal semantic scene completion for complex urban traffic scenes with reachability geometric attention
Baorong Zhan1, Xichang Yu2, Jiajun Zou1
1School of Information and Intelligence Engineering, Guangzhou Xinhua University, Dongguan, 523133, Guangdong, China.
Scientific Reports
|May 21, 2026
Summary
This study introduces AGAC, a novel network for semantic scene completion that uses reachability-aware geometric attention to accurately reconstruct occluded objects in complex urban environments. AGAC significantly improves performance in challenging scenarios like heavy occlusion and long-range perception.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Robotics
Background:
- Occlusions are a major challenge in urban autonomous driving, hindering accurate scene understanding.
- Existing methods struggle with pixel-level semantic recovery of occluded objects.
Purpose of the Study:
- To develop a novel network, AGAC, for temporally aware semantic scene completion.
- To enhance the recovery of pixel-level semantics and shapes for occluded regions.
Main Methods:
- Constructing an anisotropic reachability potential field and projecting it onto the image plane.
- Implementing reachability-aware geometric attention for guided long-range completion.
- Utilizing hierarchical occlusion parsing and a reprojection-gated memory for temporal fusion.
Main Results:
- AGAC achieves leading performance on key metrics (APQ, APC, AAP, AP_occ) on benchmarks like KITTI-360-APS and BDD100K-APS.
- Demonstrates significant improvements (3-7%) over strong baselines, especially under heavy occlusion.
- Maintains stable quality in visible regions while enhancing occluded area reconstruction.
Conclusions:
- AGAC effectively addresses the challenge of semantic scene completion in occluded urban environments.
- The proposed reachability-aware attention mechanism is crucial for accurate long-range completion.
- The method shows strong potential for improving the robustness of autonomous driving systems.
Related Concept Videos
01:15Depth Perception and Spatial Vision
Depth perception is the ability to perceive objects three-dimensionally. It relies on two types of cues: binocular and monocular. Binocular cues depend on the combination of images from both eyes and how the eyes work together. Since the eyes are in slightly different positions, each eye captures a slightly different image. This disparity between images, known as binocular disparity, helps the brain interpret depth. When the brain compares these images, it determines the distance to an object.
01:42Schemas
A schema is a mental construct consisting of a cluster or collection of related concepts (Bartlett, 1932). There are many different types of schemata, and they all have one thing in common: schemata are a method of organizing information that allows the brain to work more efficiently. When a schema is activated, the brain makes immediate assumptions about the person or object being observed.