Parameters30.5M
GFLOPs0.0
Input Size384px
Best mAP51.4%
LicenseApache-2.0
Architecture
Type
transformer
Backbone
DINOv2
Neck
Agg2Former
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 | 51.4% | 10.4 | 96.0ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 51.4% | 10.2 | 98.3ms | 150 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 51.3% | 25.4 | 39.4ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 51.4% | 23.7 | 42.1ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 51.4% | 65.7 | 15.2ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 51.4% | 39.5 | 25.3ms | 151 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 51.4% | 58.3 | 17.1ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 51.4% | 67.6 | 14.8ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
9.0ms
83.5ms
5.8ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreRFDETRn.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: 48.4% mAP