This is the code of ICPR 2018 paper《Deep Difference Analysis in Similar-looking Face Recognition》.
Using this code, you can visualize the difference in similar-looking Faces. That is, we find the most different region between two similar-looking Faces judged by the network.
- Install caffe.
- compile caffe and matcaffe (matlab wrapper for caffe)
make all -j4
make matcaffe
- Download the code.
git clone https://github.com/zhongyy/Deep-Difference-Analysis-in-Similar-looking-face-recognition.git
- Download the pretrained model and put it in the root fold.
The model could be download from Baidu Netdisk. password: dmo1 or from Google drive
We provide an example of Chow Yun Fat and a stand-in for him. Using the code, you get the result shown in paper Figure 2.
- Extract the features of Face images.
As we provide the related images and in-process data, you could run extract_zhou.m and then run generate_zhou.m to get the result from scratch.
Note that some code related to your own caffe path should be changed.
matlab extract_zhou.m
matlab generate_zhou.m
- If it is hard for you to run feature extraction code related to Matcaffe, you can run generate_zhou.m using some in-process data.
matlab generate_zhou.m
- The result should be like as follows.
- Extract the features of Face images.
As we provide the related images and in-process data, you could run extract.m and then run generate.m to get the result from scratch.
Note that some code related to your own caffe path should be changed.
matlab extract.m
matlab generate.m
- If it is hard for you to run feature extraction code related to Matcaffe, you can run generate.m using some in-process data.
matlab generate.m
- The result should be like as follows.