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

# EdgeCrafter: ECSeg

EdgeCrafter is a unified compact Vision Transformer (ViT) framework designed for efficient dense prediction tasks on edge devices. This specific checkpoint is part of the **ECSeg** series, which focuses on high-performance instance segmentation using a distilled compact backbone and an edge-friendly encoder-decoder design.

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

## Performance (Instance Segmentation on COCO2017)

| Model | Size | AP<sub>50:95</sub> | #Params | GFLOPs | Latency (ms) |
|:-----:|:----:|:--:|:-------:|:------:|:------------:|
| **ECSeg-S** | 640 | 43.0 | 10M | 33 | 6.96 |
| **ECSeg-M** | 640 | 45.2 | 20M | 64 | 9.85 |
| **ECSeg-L** | 640 | 47.1 | 34M | 111 | 12.56 |
| **ECSeg-X** | 640 | 48.4 | 50M | 168 | 14.96 |

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

## Usage

To run inference with this model, follow the instructions in the official repository. You can use the provided inference script:

```bash
# 1. Clone the repository and install dependencies
git clone https://github.com/Intellindust-AI-Lab/EdgeCrafter
cd EdgeCrafter/ecdetseg
pip install -r requirements.txt

# 2. Run PyTorch inference
# Replace `path/to/your/image.jpg` with an actual image path
python tools/inference/torch_inf.py -c configs/ecseg/ecseg_s.yml -r ecdet_s.pth -i path/to/your/image.jpg
```

## 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}
}
```

This model has been pushed to the Hub using the [PytorchModelHubMixin](https://huggingface.co/docs/huggingface_hub/package_reference/mixins#huggingface_hub.PyTorchModelHubMixin) integration.