Parameters19.0M
GFLOPs57.0
Input Size640px
Best mAP55.5%
LicenseApache-2.0
Architecture
Type
transformer
Backbone
HGNetv2
Neck
HybridEncoder
Head
DEIM (Dense O2O + MAL)
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 | 55.4% | 6.9 | 144.6ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 55.4% | 7.0 | 143.7ms | 188 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 55.5% | 23.5 | 42.5ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 55.5% | 15.7 | 63.8ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 55.5% | 57.6 | 17.4ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 55.5% | 28.4 | 35.3ms | 189 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 55.4% | 83.3 | 12.0ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 55.5% | 85.4 | 11.7ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
8.8ms
132.0ms
3.0ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreDEIMm.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: 52.7% mAP