Parameters20.0M
GFLOPs60.0
Input Size640px
Best mAP49.8%
LicenseApache-2.0
Architecture
Type
transformer
Backbone
ResNet-18
Neck
HybridEncoder
Head
DETR
Benchmark Results
Performance on COCO val2017 across different hardware configurations
| Hardware | Runtime | mAP@50-95 | FPS | Latency | VRAM |
|---|---|---|---|---|---|
| NVIDIA A100 | PyTorch FP32 | 45.6% | 27.4 | 36.5ms | 155 MB |
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 49.8% | 9.6 | 103.7ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 49.8% | 10.2 | 98.3ms | 155 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 49.7% | 38.8 | 25.7ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 49.8% | 22.1 | 45.3ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 49.8% | 92.5 | 10.8ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 49.8% | 47.2 | 21.2ms | 169 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 49.8% | 113.4 | 8.8ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 49.8% | 120.6 | 8.3ms | - |
Speed Breakdown(NVIDIA A100)
4.1ms
23.7ms
8.8ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreRTDETRr18.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: 46.5% mAP