Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,90 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- object-detection
|
| 5 |
+
- tensorrt
|
| 6 |
+
- onnx
|
| 7 |
+
- pytorch
|
| 8 |
+
- real-time
|
| 9 |
+
datasets:
|
| 10 |
+
- coco
|
| 11 |
+
library_name: transformers
|
| 12 |
+
pipeline_tag: object-detection
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# DEIMv2 - Real-Time Object Detection Meets DINOv3
|
| 16 |
+
|
| 17 |
+
Pre-trained DEIMv2 models with PyTorch checkpoints, ONNX exports, and TensorRT FP16 engines.
|
| 18 |
+
|
| 19 |
+
## Model Zoo
|
| 20 |
+
|
| 21 |
+
| Model | AP | Params | GFLOPs | Checkpoint | ONNX | TensorRT |
|
| 22 |
+
|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|
| 23 |
+
| **Atto** | 23.8 | 0.5M | 0.8 | ✅ | ✅ | ✅ |
|
| 24 |
+
| **Femto** | 31.0 | 1.0M | 1.7 | ✅ | ✅ | ✅ |
|
| 25 |
+
| **Pico** | 38.5 | 1.5M | 5.2 | ✅ | ✅ | ✅ |
|
| 26 |
+
| **N** | 43.0 | 3.6M | 6.8 | ✅ | ✅ | ✅ |
|
| 27 |
+
| **S** | 50.9 | 9.7M | 25.6 | ✅ | ✅ | ✅ |
|
| 28 |
+
| **M** | 53.0 | 18.1M | 52.2 | ✅ | ✅ | ✅ |
|
| 29 |
+
| **L** | 56.0 | 32.2M | 96.7 | ✅ | ✅ | ✅ |
|
| 30 |
+
| **X** | 57.8 | 50.3M | 151.6 | ✅ | ✅ | ✅ |
|
| 31 |
+
|
| 32 |
+
## Files
|
| 33 |
+
|
| 34 |
+
- `*.pth` - PyTorch checkpoints (EMA weights)
|
| 35 |
+
- `*.onnx` - ONNX models (opset 17, dynamic batch)
|
| 36 |
+
- `*.engine` - TensorRT FP16 engines (built on RTX 4090, TensorRT 10.14)
|
| 37 |
+
|
| 38 |
+
## Input Shapes
|
| 39 |
+
|
| 40 |
+
| Model | Input Size |
|
| 41 |
+
|:---:|:---:|
|
| 42 |
+
| Atto | 320x320 |
|
| 43 |
+
| Femto | 416x416 |
|
| 44 |
+
| Pico, N, S, M, L, X | 640x640 |
|
| 45 |
+
|
| 46 |
+
## Usage
|
| 47 |
+
|
| 48 |
+
### PyTorch
|
| 49 |
+
```python
|
| 50 |
+
from huggingface_hub import hf_hub_download
|
| 51 |
+
import torch
|
| 52 |
+
|
| 53 |
+
# Download checkpoint
|
| 54 |
+
ckpt_path = hf_hub_download("carpedm20/DEIMv2", "deimv2_dinov3_s_coco.pth")
|
| 55 |
+
checkpoint = torch.load(ckpt_path, map_location='cpu')
|
| 56 |
+
state_dict = checkpoint['ema']['module']
|
| 57 |
+
```
|
| 58 |
+
|
| 59 |
+
### ONNX Runtime
|
| 60 |
+
```python
|
| 61 |
+
import onnxruntime as ort
|
| 62 |
+
from huggingface_hub import hf_hub_download
|
| 63 |
+
|
| 64 |
+
onnx_path = hf_hub_download("carpedm20/DEIMv2", "deimv2_dinov3_s_coco.onnx")
|
| 65 |
+
session = ort.InferenceSession(onnx_path)
|
| 66 |
+
```
|
| 67 |
+
|
| 68 |
+
### TensorRT
|
| 69 |
+
```python
|
| 70 |
+
import tensorrt as trt
|
| 71 |
+
from huggingface_hub import hf_hub_download
|
| 72 |
+
|
| 73 |
+
engine_path = hf_hub_download("carpedm20/DEIMv2", "deimv2_dinov3_s_coco.engine")
|
| 74 |
+
# Load engine with TensorRT runtime
|
| 75 |
+
```
|
| 76 |
+
|
| 77 |
+
## Citation
|
| 78 |
+
|
| 79 |
+
```bibtex
|
| 80 |
+
@article{huang2025deimv2,
|
| 81 |
+
title={Real-Time Object Detection Meets DINOv3},
|
| 82 |
+
author={Huang, Shihua and Hou, Yongjie and Liu, Longfei and Yu, Xuanlong and Shen, Xi},
|
| 83 |
+
journal={arXiv},
|
| 84 |
+
year={2025}
|
| 85 |
+
}
|
| 86 |
+
```
|
| 87 |
+
|
| 88 |
+
## License
|
| 89 |
+
|
| 90 |
+
Apache 2.0 - See [DEIMv2 GitHub](https://github.com/Intellindust-AI-Lab/DEIMv2) for details.
|