Parameters25.5M
GFLOPs51.8
Input Size640px
Best mAP57.2%
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 | 57.1% | 6.4 | 156.1ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 57.1% | 6.4 | 155.6ms | 239 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 57.1% | 24.9 | 40.1ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 57.1% | 13.2 | 76.0ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 57.1% | 63.5 | 15.7ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 57.1% | 42.3 | 23.6ms | 241 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 57.2% | 74.7 | 13.4ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 57.1% | 76.7 | 13.0ms | - |
| Raspberry Pi 5 | ncnn FP32 | 55.8% | 1.9 | 531.8ms | - |
| Raspberry Pi 5 | ONNX Runtime FP32 | 55.7% | 0.7 | 1398.3ms | - |
| Raspberry Pi 5 | PyTorch FP32 | 56.5% | 0.5 | 2153.8ms | - |
| Raspberry Pi 5 + Hailo-8 | HailoRT INT8 | 54.8% | 21.2 | 47.1ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
4.6ms
144.8ms
6.2ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreYOLO9c.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: 53% mAP