DEIMv2 - Real-Time Object Detection Meets DINOv3

Pre-trained DEIMv2 models with PyTorch checkpoints, ONNX exports, and TensorRT FP16 engines.

Model Zoo

Model	AP	Params	GFLOPs	Checkpoint	ONNX	TensorRT
Atto	23.8	0.5M	0.8	✅	✅	✅
Femto	31.0	1.0M	1.7	✅	✅	✅
Pico	38.5	1.5M	5.2	✅	✅	✅
N	43.0	3.6M	6.8	✅	✅	✅
S	50.9	9.7M	25.6	✅	✅	✅
M	53.0	18.1M	52.2	✅	✅	✅
L	56.0	32.2M	96.7	✅	✅	✅
X	57.8	50.3M	151.6	✅	✅	✅

Files

• *.pth - PyTorch checkpoints (EMA weights)
• *.onnx - ONNX models (opset 17, dynamic batch)
• *.engine - TensorRT FP16 engines (built on RTX 4090, TensorRT 10.14)

Input Shapes

Model	Input Size
Atto	320x320
Femto	416x416
Pico, N, S, M, L, X	640x640

Usage

PyTorch

from huggingface_hub import hf_hub_download
import torch

# Download checkpoint
ckpt_path = hf_hub_download("carpedm20/DEIMv2", "deimv2_dinov3_s_coco.pth")
checkpoint = torch.load(ckpt_path, map_location='cpu')
state_dict = checkpoint['ema']['module']

ONNX Runtime

import onnxruntime as ort
from huggingface_hub import hf_hub_download

onnx_path = hf_hub_download("carpedm20/DEIMv2", "deimv2_dinov3_s_coco.onnx")
session = ort.InferenceSession(onnx_path)

TensorRT

import tensorrt as trt
from huggingface_hub import hf_hub_download

engine_path = hf_hub_download("carpedm20/DEIMv2", "deimv2_dinov3_s_coco.engine")
# Load engine with TensorRT runtime

Citation

@article{huang2025deimv2,
  title={Real-Time Object Detection Meets DINOv3},
  author={Huang, Shihua and Hou, Yongjie and Liu, Longfei and Yu, Xuanlong and Shen, Xi},
  journal={arXiv},
  year={2025}
}

License

Apache 2.0 - See DEIMv2 GitHub for details.