---
license: apache-2.0
pipeline_tag: image-segmentation
tags:
- model_hub_mixin
- pytorch_model_hub_mixin
---

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

EdgeCrafter is a unified framework for high-performance dense prediction on resource-constrained edge devices. It introduces compact Vision Transformers (ViTs) that compete with CNN-based architectures like YOLO by using task-specialized distillation and edge-aware encoder-decoder designs.

This repository contains a checkpoint for **ECSeg**, the instance segmentation variant of the framework, which achieves a strong accuracy-efficiency tradeoff.

- **Paper:** [EdgeCrafter: Compact ViTs for Edge Dense Prediction via Task-Specialized Distillation](https://arxiv.org/abs/2603.18739)
- **Project Page:** [EdgeCrafter Project Page](https://intellindust-ai-lab.github.io/projects/EdgeCrafter/)
- **Code:** [GitHub Repository](https://github.com/Intellindust-AI-Lab/EdgeCrafter)

## Model Description

Deploying high-performance dense prediction models on resource-constrained edge devices remains challenging due to strict limits on computation and memory. EdgeCrafter addresses this by introducing a framework centered on distilled compact backbones and edge-friendly encoder-decoder designs. For instance segmentation, ECSeg achieves performance comparable to RF-DETR while using substantially fewer parameters, proving that compact ViTs can be a practical and competitive option for edge deployment.

## Usage

This model is compatible with the `PytorchModelHubMixin`. For detailed instructions on installation, training, and running inference, please refer to the [official GitHub repository](https://github.com/Intellindust-AI-Lab/EdgeCrafter).

## Citation

If you find this project useful in your research, please consider citing:

```bibtex
@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}
}
```