Parameters18.1M
GFLOPs52.2
Input Size640px
Best mAP56.0%
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 | 56.0% | 3.8 | 260.4ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 56.0% | 4.7 | 213.5ms | 155 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 56.0% | 7.9 | 126.5ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 55.9% | 8.0 | 125.4ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 56.0% | 36.0 | 27.7ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 56.0% | 22.1 | 45.4ms | 158 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 56.0% | 49.6 | 20.2ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
17.0ms
193.4ms
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("LibreDEIMv2m.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: 53% mAP