Parameters20.0M
GFLOPs60.0
Input Size640px
Best mAP50.8%
LicenseApache-2.0
Architecture
Type
transformer
Backbone
PResNet
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 | 50.8% | 9.7 | 103.5ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 50.8% | 10.3 | 97.4ms | 155 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 50.8% | 39.9 | 25.1ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 50.7% | 22.3 | 44.8ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 50.8% | 97.1 | 10.3ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 50.8% | 51.1 | 19.6ms | 169 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 50.8% | 117.8 | 8.5ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 50.7% | 125.5 | 8.0ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
4.5ms
90.0ms
2.9ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreRTDETRv2r18.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: 48.1% mAP