Publications
Our published research in top-tier journals and conferences.
Real-Time Object Detection in Aerial Imagery Using Lightweight Transformers
Computer Vision and Pattern Recognition (CVPR), 2025
This paper presents a lightweight transformer-based architecture for real-time object detection in UAV-captured aerial images. Our model achieves 89.7% mAP on the VisDrone dataset while maintaining inference speeds above 45 FPS on edge devices. We in...
Efficient Panoptic Segmentation for Autonomous Driving Scenes
International Conference on Computer Vision (ICCV), 2025
We propose an efficient panoptic segmentation model designed for autonomous driving applications. Our architecture employs a shared backbone with dual decoders for simultaneous semantic and instance segmentation, achieving 65.2 PQ on the Cityscapes b...
Temporal Action Detection in Surveillance Videos Using Graph Neural Networks
European Conference on Computer Vision (ECCV), 2024
This paper introduces a graph neural network-based framework for temporal action detection in surveillance videos. We model the spatio-temporal relationships between human body keypoints as dynamic graphs and employ a message-passing mechanism for ac...