Parameters2.1M
GFLOPs2.5
Input Size416px
Best mAP37.9%
LicenseApache-2.0
Architecture
Type
one-stage
Backbone
ESNet
Neck
CSP-PAN
Head
GFL
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 | 37.3% | 10.7 | 93.4ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 37.9% | 12.5 | 80.3ms | 26 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 37.9% | 32.4 | 30.8ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 37.9% | 29.6 | 33.7ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 37.9% | 46.0 | 21.7ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 37.9% | 45.1 | 22.2ms | 38 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 37.3% | 65.7 | 15.2ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 36.0% | 53.5 | 18.7ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
5.7ms
53.0ms
21.6ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibrePICODETm.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: 34.3% mAP