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

# EdgeCrafter: ECSeg-L

EdgeCrafter is a unified compact Vision Transformer (ViT) framework designed for efficient edge dense prediction. This specific model, **ECSeg-L**, is optimized for instance segmentation on resource-constrained devices. It is part of the work presented in [EdgeCrafter: Compact ViTs for Edge Dense Prediction via Task-Specialized Distillation](https://arxiv.org/abs/2603.18739).

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

## Model Description
EdgeCrafter addresses the performance gap between compact ViTs and CNN-based architectures like YOLO on edge devices. By using task-specialized distillation and an edge-friendly encoder-decoder design, EdgeCrafter models achieve a strong accuracy-efficiency tradeoff. ECSeg-L provides a high-performance balance for instance segmentation tasks.

## Usage
To use this model, please refer to the [official GitHub repository](https://github.com/Intellindust-AI-Lab/EdgeCrafter) for installation instructions. You can run inference using the following command:

```bash
cd ecdetseg
# Run PyTorch inference
# Make sure to replace `path/to/your/image.jpg` with an actual image path and provide the path to the weights
python tools/inference/torch_inf.py -c configs/ecseg/ecseg_l.yml -r /path/to/ecseg_l.pth -i path/to/your/image.jpg
```

For loading models directly via the Hugging Face Hub, check the [hf_models.ipynb](https://github.com/Intellindust-AI-Lab/EdgeCrafter/blob/main/hf_models.ipynb) notebook in the repository.

## Citation
```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}
}
```