Parameters19.0M
GFLOPs57.0
Input Size640px
Best mAP57.9%
LicenseApache-2.0
Architecture
Type
transformer
Backbone
HGNetv2
Neck
HybridEncoder
Head
FDR (Fine-grained Distribution Refinement)
Benchmark Results
Performance on COCO val2017 across different hardware configurations
| Hardware | Runtime | mAP@50-95 | FPS | Latency | VRAM |
|---|---|---|---|---|---|
| NVIDIA A100 | PyTorch FP32 | 55.1% | 21.7 | 46.1ms | 187 MB |
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 57.8% | 6.9 | 144.4ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 57.8% | 7.0 | 143.3ms | 188 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 57.9% | 23.7 | 42.3ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 57.7% | 15.8 | 63.5ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 57.8% | 59.8 | 16.7ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 57.8% | 28.8 | 34.8ms | 189 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 57.6% | 74.5 | 13.4ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 57.8% | 85.3 | 11.7ms | - |
Speed Breakdown(NVIDIA A100)
6.3ms
29.8ms
10.0ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreDFINEm.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,)detrnms-freePaper: 52.3% mAP