Parameters42.0M
GFLOPs136.0
Input Size640px
Best mAP55.9%
LicenseApache-2.0
Architecture
Type
transformer
Backbone
ResNet-50
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 | 52.7% | 22.7 | 44.1ms | 322 MB |
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 55.9% | 4.4 | 228.0ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 55.9% | 4.3 | 234.6ms | 323 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 55.7% | 23.6 | 42.4ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 55.9% | 12.6 | 79.1ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 55.9% | 62.8 | 15.9ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 55.9% | 29.8 | 33.6ms | 326 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 55.8% | 97.4 | 10.3ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 55.9% | 87.4 | 11.4ms | - |
Speed Breakdown(NVIDIA A100)
3.4ms
31.8ms
9.0ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreRTDETRr50.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: 53.1% mAP