Parameters32.2M
GFLOPs96.3
Input Size640px
Best mAP58.6%
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 | 58.5% | 2.7 | 370.2ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 58.6% | 3.3 | 303.8ms | 204 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 58.5% | 6.2 | 160.8ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 58.6% | 6.1 | 163.3ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 58.6% | 31.4 | 31.9ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 58.6% | 19.7 | 50.7ms | 209 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 58.5% | 38.7 | 25.9ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 58.6% | 43.2 | 23.2ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
17.0ms
283.8ms
3.1ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreDEIMv2l.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: 56% mAP