Parameters20.1M
GFLOPs38.7
Input Size640px
Best mAP56.1%
LicenseMIT
Architecture
Type
one-stage
Backbone
GELAN
Neck
PGI
Head
Decoupled
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 | 56.1% | 7.8 | 128.1ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 56.1% | 8.0 | 124.8ms | 183 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 56.1% | 25.8 | 38.8ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 56.1% | 14.5 | 68.9ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 56.1% | 64.5 | 15.5ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 56.1% | 37.1 | 26.9ms | 184 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 56.1% | 72.3 | 13.8ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 56.1% | 77.5 | 12.9ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
4.6ms
114.3ms
6.0ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreYOLO9m.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,)anchor-freenmsPaper: 51.4% mAP