Parameters3.6M
GFLOPs6.9
Input Size640px
Best mAP46.7%
LicenseApache-2.0
Architecture
Type
transformer
Backbone
DINOv3-distilled ViT
Neck
HybridEncoder
Head
DEIM
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 | 46.7% | 17.6 | 56.8ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 46.7% | 11.1 | 89.8ms | 63 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 46.6% | 41.2 | 24.3ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 46.7% | 33.0 | 30.3ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 46.7% | 76.3 | 13.1ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 46.7% | 35.1 | 28.5ms | 70 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 46.6% | 78.0 | 12.8ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 46.7% | 70.1 | 14.3ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
9.5ms
77.4ms
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("LibreDEIMv2n.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: 43% mAP