zhaofang0627 / face-deocc-lstm

Robust LSTM-Autoencoders for Face De-Occlusion in the Wild

Robust LSTM-Autoencoders for Face De-Occlusion in the Wild

Code for paper Robust LSTM-Autoencoders for Face De-Occlusion in the Wild by Fang Zhao, Jiashi Feng, Jian Zhao, Wenhan Yang, Shuicheng Yan; TIP 2018.

Getting Started

To compile cudamat library, modify CUDA_ROOT in cudamat/Makefile to the relevant cuda root path.

Install caffe and pycaffe.

Next compile .proto file by calling

protoc -I=./ --python_out=./ config.proto

Training and Test

lstm_ae_spatial_mr_com.py: training and test for the RLA model described in the paper.

lstm_ae_spatial_mr_com_ladv.py: training and test for the Identity Preserving RLA (IP-RLA) model described in the paper.

face-deocc-lstm's Issues

The dataset

Will you release the synthetic dataset your paper used? And the generate script?

How can I train this model with my own dataset?

Hello, I study Face Recognition and I'm interested in your paper.

First of all, Thank you for sharing your code. But, I have a problem with using your source code. The problem is that I don't know how to train this source code with my own dataset.

Could you let me know the way to train my own dataset?

Thank you.

what tools are used to generate occluded faces, can it be released?

I have some question in this paper(Robust LSTM-Autoencoders for Face De-Occlusion in the Wild),
what tools did the author use to generate occluded faces? Can it be released?
And when will the database be released?

Error: no module named 'gpu_lock2'

Hi. Thanks for releasing the source code. When I try to run the code I get the error "no module named 'gpu_lock2'" under the util.py file. I googled this but found no such open source library. How to fix this?