license: apache-2.0
pipeline_tag: object-detection
tags:
- model_hub_mixin
- pytorch_model_hub_mixin
EdgeCrafter: Compact ViTs for Edge Dense Prediction
EdgeCrafter is a unified framework for compact Vision Transformers (ViTs) optimized for edge dense prediction tasks like object detection, instance segmentation, and pose estimation. It bridges the performance gap between compact ViTs and CNN-based architectures (like YOLO) through task-specialized distillation and edge-friendly architectural designs.
- Paper: EdgeCrafter: Compact ViTs for Edge Dense Prediction via Task-Specialized Distillation
- Project Page: https://intellindust-ai-lab.github.io/projects/EdgeCrafter/
- Repository: https://github.com/Intellindust-AI-Lab/EdgeCrafter
Model Description
Compact ViTs often struggle with task-specific representation learning on edge devices. EdgeCrafter addresses this by introducing ECDet, a detection model built from a distilled compact backbone. On the COCO dataset, ECDet-S achieves 51.7 AP with fewer than 10M parameters. This framework demonstrates that when paired with task-specialized distillation, compact ViTs are highly competitive for real-time edge applications.
Quick Start (Inference)
To run inference on a sample image using the pre-trained model, follow these steps from the official repository:
# 1. Clone the repository and install dependencies
git clone https://github.com/Intellindust-AI-Lab/EdgeCrafter
cd EdgeCrafter
pip install -r requirements.txt
# 2. Run PyTorch inference
# Make sure to replace `path/to/your/image.jpg` with an actual image path
cd ecdetseg
python tools/inference/torch_inf.py -c configs/ecdet/ecdet_l.yml -r ecdet_l.pth -i path/to/your/image.jpg
Citation
If you find this project useful in your research, please consider citing:
@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}
}