Parameters32.1M
GFLOPs0.0
Input Size512px
Best mAP55.3%
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 | 55.1% | 6.2 | 161.5ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 55.1% | 6.9 | 144.6ms | 175 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 55.2% | 15.5 | 64.4ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 55.1% | 14.5 | 69.1ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 55.1% | 68.5 | 14.6ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 55.1% | 35.8 | 27.9ms | 176 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 55.3% | 71.6 | 14.0ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 55.1% | 77.5 | 12.9ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
13.1ms
125.8ms
5.7ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreRFDETRs.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