Skip to content

README_zh.md

Latest commit

 

History

History
139 lines (100 loc) · 7.76 KB

README_zh.md

File metadata and controls

139 lines (100 loc) · 7.76 KB
Raw

EdgeCrafter: Compact ViTs for Edge Dense Prediction via Task-Specialized Distillation

English | 简体中文

Webpage arXiv License

Longfei Liu *‡  Yongjie Hou *  Yang Li *  Qirui Wang *  Youyang Sha 
Yongjun Yu  Yinzhi Wang  Peizhe Ru  Xuanlong Yu†  Xi Shen †

* Equal Contribution    ‡ Project Leader    † Corresponding Author

Intellindust AI Lab

🚀 更新日志

🤗 Hugging Face

模型已在 🤗 Hugging Face 开放下载！也可以通过 hf_models.ipynb 快速调用模型。欢迎尝试！

📍 结果复现

🏆 模型库

COCO2017 Validation Results

Note: Latency is measured on an NVIDIA T4 GPU with batch size 1 under FP16 precision using TensorRT (v10.6).

Object Detection

Model Size AP50:95 #Params GFLOPs Latency (ms) Config Log Checkpoint
ECDet-S 640 51.7 10 26 5.41 config log model
ECDet-M 640 54.3 18 53 7.98 config log model
ECDet-L 640 57.0 31 101 10.49 config log model
ECDet-X 640 57.9 49 151 12.70 config log model

Instance Segmentation

Model Size AP50:95 #Params GFLOPs Latency (ms) Config Log Checkpoint
ECSeg-S 640 43.0 10 33 6.96 config log model
ECSeg-M 640 45.2 20 64 9.85 config log model
ECSeg-L 640 47.1 34 111 12.56 config log model
ECSeg-X 640 48.4 50 168 14.96 config log model

Pose Estimation

Model Size AP50:95 #Params GFLOPs Latency (ms) Config Log Checkpoint
ECPose-S 640 68.9 10 30 5.54 config log model
ECPose-M 640 72.4 20 63 9.25 config log model
ECPose-L 640 73.5 34 112 11.83 config log model
ECPose-X 640 74.8 51 172 14.31 config log model

📦 安装

# 创建并激活 conda 环境
conda create -n ec python=3.11 -y
conda activate ec

# 安装依赖
pip install -r requirements.txt

⚡ 快速上手（模型推理）

可以通过预训练模型对示例图像进行推理，以快速测试 EdgeCrafter 的性能。

# 1. 进入对应目录并下载预训练权重（以 ECDet-L 为例）
cd ecdetseg
wget https://github.com/capsule2077/edgecrafter/releases/download/edgecrafterv1/ecdet_l.pth

# 2. 运行 PyTorch 推理
# 请将 `path/to/your/image.jpg` 替换为实际图像的路径
python tools/inference/torch_inf.py -c configs/ecdet/ecdet_l.yml -r ecdet_l.pth -i path/to/your/image.jpg

📄 开源协议

本项目遵循 Apache 2.0 许可证 开源。

🙏 致谢

感谢以下开源项目为本工作提供的支持与启发：RT-DETR、D-FINE、DEIM、lightly-train、DETRPose、RF-DETR、DINOv3

📚 引用

如果您在研究中使用了本项目，请引用：

@article{liu2026edgecrafter,
  title={EdgeCrafter: Compact ViTs for Edge Dense Prediction via Task-Specialized Distillation},
  author={Liu, Longfei and Hou, Yongjie and Li, Yang and Wang, Qirui and Sha, Youyang and Yu, Yongjun and Wang, Yinzhi and Ru, Peizhe and Yu, Xuanlong and Shen, Xi},
  journal={arXiv},
  year={2026}
}