Parameters1.5M
GFLOPs5.2
Input Size640px
Best mAP42.3%
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 | 42.3% | 21.3 | 47.1ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 42.2% | 14.1 | 71.1ms | 55 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 42.3% | 44.5 | 22.5ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 42.3% | 37.7 | 26.5ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 42.3% | 80.2 | 12.5ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 42.3% | 40.5 | 24.7ms | 62 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 42.2% | 79.9 | 12.5ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
9.4ms
59.0ms
2.7ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreDEIMv2pico.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: 38.5% mAP