| Property | Value |
|---|---|
| Parameter Count | 20M parameters |
| Model Type | Real-time Object Detection |
| License | Apache-2.0 |
| Paper | arXiv:2304.08069 |
What is rtdetr_r18vd?
RT-DETR R18VD is a groundbreaking real-time object detection model that bridges the gap between YOLO-style detectors and Transformer-based approaches. It achieves 46.5% AP on COCO while maintaining an impressive 217 FPS on T4 GPU, making it one of the fastest end-to-end object detectors available.
Implementation Details
The model employs a novel architecture combining an efficient hybrid encoder with uncertainty-minimal query selection. It processes multi-scale features through Attention-based Intra-scale Feature Interaction (AIFI) and CNN-based Cross-scale Feature Fusion (CCFF), optimizing both speed and accuracy.
- Efficient hybrid encoder for fast multi-scale feature processing
- Uncertainty-minimal query selection for high-quality initial queries
- Flexible speed tuning through adjustable decoder layers
- End-to-end detection without NMS (Non-Maximum Suppression)
Core Capabilities
- Real-time object detection at 217 FPS
- 46.5% AP on COCO val2017
- 63.8% AP50 and 50.4% AP75
- Effective across different object scales (APs: 28.4%, APm: 49.8%, APl: 63.0%)
Frequently Asked Questions
Q: What makes this model unique?
RT-DETR R18VD is the first real-time end-to-end object detector that eliminates the need for NMS while maintaining both high speed and accuracy. It uniquely combines Transformer-based architecture with efficient processing techniques.
Q: What are the recommended use cases?
The model is ideal for real-time object detection applications where both speed and accuracy are crucial, such as surveillance systems, autonomous vehicles, and real-time video analytics. Its flexible speed tuning makes it adaptable to various deployment scenarios.