Parameters67.0M
GFLOPs234.0
Input Size640px
Best mAP58.0%
LicenseApache-2.0
Architecture
Type
transformer
Backbone
HGNetv2-X
Neck
HybridEncoder
Head
DETR
Benchmark Results
Performance on COCO val2017 across different hardware configurations
| Hardware | Runtime | mAP@50-95 | FPS | Latency | VRAM |
|---|---|---|---|---|---|
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 58.0% | 3.0 | 331.4ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 57.9% | 3.0 | 338.1ms | 394 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 57.9% | 17.0 | 58.9ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 58.0% | 8.6 | 115.8ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 58.0% | 48.4 | 20.7ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 57.9% | 21.8 | 46.0ms | 401 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 57.9% | 82.3 | 12.2ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 57.9% | 68.5 | 14.6ms | - |
| Raspberry Pi 5 | ONNX Runtime FP32 | 58.0% | 0.4 | 2801.5ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
4.5ms
330.9ms
2.7ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreRTDETRx.pt")
# Run inference
result = model("image.jpg", conf=0.25, iou=0.45)
# Process results
print(f"Found {len(result)} objects")
print(result.boxes.xyxy) # bounding boxes (N, 4)
print(result.boxes.conf) # confidence scores (N,)
print(result.boxes.cls) # class IDs (N,)detrnms-freePaper: 54.8% mAP