Parameters19.0M
GFLOPs57.0
Input Size640px
Best mAP56.5%
LicenseApache-2.0
Architecture
Type
transformer
Backbone
HGNetv2
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 | 56.5% | 6.7 | 149.1ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 56.5% | 7.0 | 143.1ms | 188 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 56.5% | 23.1 | 43.3ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 56.4% | 15.4 | 65.0ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 56.5% | 56.6 | 17.7ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 56.5% | 33.6 | 29.8ms | 189 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 56.4% | 86.7 | 11.5ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 56.5% | 84.4 | 11.9ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
8.8ms
131.4ms
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("LibreRTDETRv4m.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: 53.5% mAP