Parameters1.0M
GFLOPs1.7
Input Size416px
Best mAP34.5%
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 | 34.5% | 37.1 | 26.9ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 34.5% | 14.8 | 67.5ms | 29 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 34.5% | 65.7 | 15.2ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 34.5% | 60.0 | 16.7ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 34.5% | 90.3 | 11.1ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 34.5% | 45.3 | 22.1ms | 33 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 34.3% | 119.0 | 8.4ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
7.1ms
57.6ms
2.8ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import LibreYOLO
# Load model (auto-downloads from HuggingFace if not found locally)
model = LibreYOLO("LibreDEIMv2femto.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: 31% mAP