Thanks to zeakey's help. Created by XuanyiLi, if you have any problem in using it, please contact:[email protected]. The best result of my pytorch model is 0.772 ODS F-score now.
the following are the side outputs and the prediction example
SGD no tunelr 1e-8:
Adam no tunelr 1e-4:
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},
}
I implement the edge detection model according to the HED 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 |
Install pytorch. 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.
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
Side Note: Hello mingyang, I love you
Our code is released under MIT License (see LICENSE file for details).
- 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.
*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)
[1] HED