Parameters99.1M
GFLOPs141.2
Input Size640px
Best mAP56.3%
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 | 56.3% | 3.6 | 273.8ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 56.3% | 3.5 | 289.6ms | 505 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 56.2% | 14.6 | 68.7ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 56.3% | 6.7 | 149.3ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 56.3% | 41.6 | 24.0ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 56.3% | 39.6 | 25.2ms | 511 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 56.3% | 72.1 | 13.9ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 56.3% | 48.4 | 20.7ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
8.9ms
272.3ms
8.5ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreYOLOXx.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: 51.1% mAP