sampson-lee / mydrgan

Pytorch implimentation of《2017CVPR-Representation Learning by Rotating Your Faces》

I'm Xinpeng (Sampson) Li.

About me

• 🎓 A PhD candidate at CS of UTD, advised by Yapeng Tian. Here is my CV.
• 👀 Currently engaged in artificial intelligence and social interaction research.
• 📫 Feel free to reach out at [email protected].

Publications

Two in One Go: Single-stage Emotion Recognition with Decoupled Subject-context Transformer.
Xinpeng Li, Teng Wang, Jian Zhao, Shuyi Mao, Jinbao Wang, Feng Zheng, Xiaojiang Peng, Xuelong Li.
Paper & Code will come soon.
Accepted by ACMMM'24.

Real3D-AD: A Dataset of Point Cloud Anomaly Detection.
Jiaqi Liu, Guoyang Xie, Ruitao Chen, Xinpeng Li, Jinbao Wang, Yong Liu, Chengjie Wang, Feng Zheng.
Paper & Code.
Accepted by NeurIPS'23.

Facial Action Units as A Bridge of Joint Dataset Training for Facial Expression Recognition.
Shuyi Mao, Xinpeng Li, Fan Zhang, Xiaojiang Peng, and Yang Yang.
Paper (Preliminary version) & Code will come soon.
Submitted to TMM'23.

Rail Detection: An Efficient Row-based Network and a New Benchmark.
Xinpeng Li, and Xiaojiang Peng.
Paper & Code.
Accepted by ACMMM'22.

Sequential Interactive Biased Network for Context-Aware Emotion Recognition.
Xinpeng Li, Xiaojiang Peng, and Changxing Ding.
Paper & Code.
Accepted by IJCB'21.

mydrgan's Issues

What does 'enable the reconstruct loss' do?

Hello Sampson,
I saw in train_single_DRGAN.py you add
if args.use_rec: g_loss += 10*L_g_rec
Could you please tell me what it does for?

When I were training with enabling the reconstruct loss, after many epochs(about 50 epochs), the genbySimage seemed good. However, when I used the model to generate images, the faces were very blurred.

where is the multi_DR_GAN_list.txt?

Hello, where is the multi_DR_GAN_list.txt? In \data, there is only single_DR_GAN_list.txt

bug report

line here I think it should be:

     batch_pose_code = torch.FloatTensor(batch_pose_code.tolist())

Can you share your trained model Thanks!

Hello Sampson,
I think your implementation is great and inspiring.I saw you update the code, could you share your new trained model?
Thanks a lot!

multipie DATASET

where can i and in which which format i can upload multipie dataset to train my own model

could you share your model again？Thanks

A question about testing the goodgen_singel_G.pth

Thanks for your sharing.
I have tested the model you have shared.But I found some wrong.I feed 30 images to the generator, and all the images are different person, but the result is all the generated images are same.I don't know what happend. Can you help me ?

About the dataset.

Hello Sampson,
could you please me some details about the dataset you used when you trained the model?

Python version and the versions of the dependencies

Hello, I noticed that the README.md does not specify the required Python version and the versions of the dependencies.

Could you please provide the following information:

• Recommended Python version for this project
• Specific versions of the dependencies that are known to work with this project (if applicable)

Thank you for your help!

Could you share you model again?

I thought your implementation is great but the model link in Baidu expired...Could you upload it again please?

How do you test your generated images(Multi-pie dataset)?

In paper, the first 200 subjects are for training while the rest 137 for testing. For testing,
one image per subject with the frontal view and neutral illumination is the gallery and the others are the probes.
But I don't understand how you transform your test images to feature vectors? Do you use the discriminator or another face recognition method?