Parameters25.3M
GFLOPs37.0
Input Size640px
Best mAP51.8%
LicenseApache-2.0
Architecture
Type
one-stage
Backbone
CSPDarknet
Neck
PAFPN
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 | 51.8% | 9.3 | 107.3ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 51.7% | 9.1 | 110.1ms | 172 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 51.8% | 28.0 | 35.7ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 51.8% | 16.4 | 61.1ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 51.8% | 55.8 | 17.9ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 51.7% | 48.2 | 20.7ms | 172 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 51.6% | 83.0 | 12.0ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 51.7% | 70.1 | 14.3ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
8.8ms
92.8ms
8.6ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreYOLOXm.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: 46.9% mAP