Parameters36.6M
GFLOPs0.0
Input Size640px
Best mAP53.9%
LicenseApache-2.0
Architecture
Type
transformer
Backbone
ResNet-50
Neck
HybridEncoder (0.5 expansion)
Head
DETR (eval at decoder layer 3 of 6)
Benchmark Results
Performance on COCO val2017 across different hardware configurations
| Hardware | Runtime | mAP@50-95 | FPS | Latency | VRAM |
|---|---|---|---|---|---|
| NVIDIA A100 | PyTorch FP32 | 50.8% | 25.1 | 39.8ms | 298 MB |
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 53.8% | 5.4 | 184.5ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 53.8% | 5.0 | 198.3ms | 298 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 53.8% | 29.1 | 34.3ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 53.8% | 15.7 | 63.6ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 53.8% | 76.4 | 13.1ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 53.8% | 35.3 | 28.4ms | 302 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 53.9% | 110.1 | 9.1ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 53.9% | 102.8 | 9.7ms | - |
Speed Breakdown(NVIDIA A100)
3.2ms
27.5ms
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("LibreRTDETRr50m.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-free