Parameters2.0M
GFLOPs4.0
Input Size640px
Best mAP41.8%
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 | 41.8% | 18.6 | 53.8ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 41.8% | 9.9 | 101.2ms | 47 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 41.8% | 34.3 | 29.1ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 41.8% | 27.8 | 36.0ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 41.8% | 52.1 | 19.2ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 41.8% | 31.9 | 31.3ms | 48 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 41.8% | 59.0 | 17.0ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 41.8% | 72.0 | 13.9ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
4.6ms
89.3ms
7.4ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreYOLO9t.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: 38.3% mAP