RT-DETR R101VD COCO-Objects365
| Property | Value |
|---|---|
| Parameter Count | 76.8M |
| License | Apache-2.0 |
| Paper | arXiv:2304.08069 |
| Performance | 56.2% AP on COCO, 74 FPS on T4 GPU |
What is rtdetr_r101vd_coco_o365?
RT-DETR R101VD is a state-of-the-art real-time object detection model that combines the accuracy of DETR (Detection Transformer) architectures with the speed typically associated with YOLO models. This particular variant uses a ResNet101 backbone and has been trained on both COCO and Objects365 datasets, achieving impressive results of 56.2% AP while maintaining real-time performance at 74 FPS.
Implementation Details
The model implements an efficient hybrid encoder architecture that processes multi-scale features through two key components: Attention-based Intra-scale Feature Interaction (AIFI) and CNN-based Cross-scale Feature Fusion (CCFF). It uses an innovative uncertainty-minimal query selection method to optimize initial object queries for the decoder.
- Efficient hybrid encoder for multi-scale feature processing
- Uncertainty-minimal query selection for improved accuracy
- Flexible speed tuning through adjustable decoder layers
- Pre-trained on Objects365 for enhanced performance
Core Capabilities
- Real-time object detection at 74 FPS on T4 GPU
- High accuracy with 56.2% AP on COCO dataset
- Excellent performance across different object scales (AP-s: 38.3%, AP-m: 60.5%, AP-l: 73.5%)
- Support for 80 COCO object categories
Frequently Asked Questions
Q: What makes this model unique?
This model uniquely bridges the gap between DETR and YOLO architectures, offering both high accuracy and real-time performance without the need for Non-Maximum Suppression (NMS). It's the first end-to-end object detector to achieve this balance.
Q: What are the recommended use cases?
The model is ideal for real-time object detection applications that require both speed and accuracy, such as autonomous driving, surveillance systems, and industrial inspection. Its flexible speed tuning makes it adaptable to various hardware configurations.
