Parameters9.7M
GFLOPs25.6
Input Size640px
Best mAP53.0%
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 | 53.0% | 5.1 | 194.4ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 53.0% | 5.7 | 175.8ms | 101 MB |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 53.0% | 10.5 | 95.5ms | - |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP32 | 53.0% | 10.4 | 95.7ms | - |
| NVIDIA RTX 5070 Ti | ONNX Runtime FP32 | 53.0% | 39.6 | 25.2ms | - |
| NVIDIA RTX 5070 Ti | PyTorch FP32 | 53.0% | 24.3 | 41.2ms | 104 MB |
| NVIDIA RTX 5070 Ti | TensorRT FP16 | 53.0% | 50.4 | 19.8ms | - |
| NVIDIA RTX 5070 Ti | TensorRT FP32 | 53.0% | 52.5 | 19.0ms | - |
Speed Breakdown(NVIDIA Jetson Orin Nano Super 8GB)
17.0ms
155.9ms
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("LibreDEIMv2s.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: 50.9% mAP