Parameters31.0M
GFLOPs91.0
Input Size640px
Best mAP57.8%
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 | 57.8% | 4.8 | 206.5ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 57.8% | 4.7 | 210.8ms | 246 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 57.8% | 18.1 | 55.2ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 57.8% | 12.1 | 82.4ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 57.8% | 48.0 | 20.8ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 57.8% | 24.7 | 40.4ms | 245 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 57.7% | 78.0 | 12.8ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 57.8% | 72.7 | 13.8ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
8.8ms
199.0ms
3.0ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreRTDETRv4l.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: 55.4% mAP