This code is update thecode into python3,and add the re_Dice loss which mentioned in Learning to Predict Crisp Boundaries(ECCV 2018).we apply it into face edge detection task,the experiment has achieved good results.
The first paper proposes a Bi-Directional Cascade Network for edge detection. By introducing a bi-directional cascade structure to enforce each layer to focus on a specific scale, BDCN trains each network layer with a layer-specific supervision. To enrich the multi-scale representations learned with a shallow network, we further introduce a Scale Enhancement Module (SEM). Here are the code for this paper.
The second paper mainly proposed a useful loss to help the network learning a crisp edge .
(left:the original results,right:add the re_Dice loss results)
- pytorch >= 0.2.0(Our code is based on the 0.2.0)
- numpy >= 1.11.0
- pillow >= 3.3.0
- python3
- Clone this repository to local
git clone https://github.com/pytorch/pytorch.git
-
Download the imagenet pretrained vgg16 pytorch model [vgg16.pth](link: https://pan.baidu.com/s/10Tgjs7FiAYWjVyVgvEM0mA code: ab4g) or the caffemodel from the model zoo and then transfer to pytorch version. You also can download our pretrained model for only evaluation. The google drive link.
-
Download the dataset to the local folder
-
running the training code train.py or test code test.py
BDCN model for BSDS500 dataset and NYUDv2 datset of RGB and depth are availavble on Baidu Disk.
The link https://pan.baidu.com/s/18PcPQTASHKD1-fb1JTzIaQ
code: j3de
The pretrained model will be updated soon.