R

Rtdetr R50vd

Developed by PekingU
RT-DETR is the first real-time end-to-end object detector, achieving a balance of speed and accuracy surpassing the YOLO series through efficient hybrid encoders and query selection mechanisms
Downloads 29.40k
Release Time : 5/29/2024

Model Overview

Transformer-based real-time object detection model that eliminates the Non-Maximum Suppression (NMS) step in traditional detectors, supporting flexible speed adjustment by tuning the number of decoder layers

Model Features

Real-time End-to-end Detection
The first real-time detector eliminating NMS, avoiding the negative impact of NMS on speed and accuracy
Efficient Hybrid Encoder
Rapidly processes multi-scale features by decoupling intra-scale interaction and cross-scale fusion
Flexible Speed Adjustment
Supports flexible switching between different speed levels by adjusting the number of decoder layers
Uncertainty Minimization Queries
Innovative query selection mechanism provides high-quality initial queries for the decoder

Model Capabilities

Real-time Object Detection
Multi-scale Feature Processing
End-to-end Prediction
COCO Dataset Object Recognition

Use Cases

Smart Surveillance
Public Space Crowd Monitoring
Real-time detection of pedestrians and luggage in scenarios like airports and train stations
High frame-rate processing capability at 108FPS@T4 GPU
Autonomous Driving
Road Object Recognition
Real-time detection of vehicles, pedestrians, traffic signs, etc.
54.3% AP@RT-DETR-R101
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase