Parameters32.0M
GFLOPs110.0
Input Size640px
Best mAP55.8%
LicenseApache-2.0
Architecture
Type
transformer
Backbone
HGNetv2-L
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 | 55.8% | 4.8 | 206.5ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 55.8% | 4.9 | 206.0ms | 230 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 55.8% | 24.1 | 41.5ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 55.8% | 13.6 | 73.5ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 55.8% | 64.5 | 15.5ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 55.8% | 27.7 | 36.1ms | 231 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 55.8% | 122.0 | 8.2ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 55.7% | 94.2 | 10.6ms | - |
| Raspberry Pi 5 | ONNX Runtime FP32 | 55.8% | 0.7 | 1461.0ms | - |
| Raspberry Pi 5 | PyTorch FP32 | 55.8% | 0.4 | 2345.2ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
4.5ms
198.8ms
2.8ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreRTDETRl.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: 53% mAP