Awesome

pytorch code for Holistically-Nested Edge Detection

Thanks to <a href="https://github.com/zeakey">zeakey's</a> help. Created by XuanyiLi, if you have any problem in using it, please contact:xuanyili.edu@gmail.com. The best result of my pytorch model is 0.772 ODS F-score now.

my model result

the following are the side outputs and the prediction example SGD no tunelr 1e-8: Adam no tunelr 1e-4:

Citation

If you find our work useful in your research, please consider citing:

    @InProceedings{xie15hed,
    author = {"Xie, Saining and Tu, Zhuowen"},
    Title = {Holistically-Nested Edge Detection},
    Booktitle = "Proceedings of IEEE International Conference on Computer Vision",
    Year  = {2015},
    }

Introduction

I implement the edge detection model according to the <a href="https://github.com/s9xie/hed">HED</a> model in pytorch.

the result of my pytorch model will be released in the future

Method	ODS F-score on BSDS500 dataset	ODS F-score on NYU Depth dataset
Ours	0.772	***
Refere nce[1]	0.782/0.789	0.746

Installation

Install <a href="https://pytorch.org/">pytorch</a>. The code is tested under 0.4.1 GPU version and Python 3.6 on Ubuntu 16.04. There are also some dependencies for a few Python libraries for data processing and visualizations like cv2 etc. It's highly recommended that you have access to GPUs.

Usage

image edge detection

To train a HED model on BSDS500:

    python train_hed.py

If you have multiple GPUs on your machine, you can also run the multi-GPU version training:

    CUDA_VISIBLE_DEVICES=0,1 python train_multi_gpu.py --num_gpus 2

After training, to evaluate:

    python evaluate.py

<i>Side Note:</i> Hello mingyang, I love you

License

Our code is released under MIT License (see LICENSE file for details).

Updates

To do

Add support for multi-gpu training for the edge detetion task.
Improve the performance to 0.782 in the original paper.
Add a gpu version of edge-eval code to accelerate the evaluation process.

source

*To download the pretrained model, please click https://drive.google.com/open?id=1nvmTv69lpXOHbqTWQLY5nRzhGR7MTBrg. (This pretrained model is not the best model, just for communicating)

Related Projects

[1] <a href="https://github.com/s9xie/hed">HED</a>

[2] <a href="https://github.com/zeakey/hed">HED</a> created by <a href="https://github.com/zeakey">zeakey's</a>