Parameters62.0M
GFLOPs202.0
Input Size640px
Best mAP61.5%
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 | 59.3% | 15.4 | 64.7ms | 401 MB |
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 61.4% | 3.0 | 328.9ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 61.4% | 2.9 | 350.0ms | 399 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 61.4% | 14.2 | 70.5ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 61.5% | 8.0 | 124.8ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 61.4% | 41.5 | 24.1ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 61.4% | 18.1 | 55.1ms | 404 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 61.5% | 73.0 | 13.7ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 61.4% | 57.9 | 17.3ms | - |
Speed Breakdown(NVIDIA A100)
6.4ms
48.6ms
9.8ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreDFINEx.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: 55.8% mAP