jixinya / eamm Goto Github PK

Hello! Thank you for sharing the code. Could you please also share dataset process? I want to train your model from scratch. Thank you

run and error:

(EAMM) root@ubuntu-desktop:~/EAMM# python demo.py
Traceback (most recent call last):
  File "demo.py", line 44, in <module>
    predictor = dlib.shape_predictor('./shape_predictor_68_face_landmarks.dat')
RuntimeError: Unable to open ./shape_predictor_68_face_landmarks.dat

work directory:

(EAMM) root@ubuntu-desktop:~/EAMM# ls
3DDFA_V2         demo               LICENSE            modules          README.md         %SystemDrive%  venv
augmentation.py  demo.py            log                ops.py           requirements.txt  teaser-1.png
config           filter1.py         logger.py          process_data.py  run.py            test
dataset          frames_dataset.py  M003_template.npy  __pycache__      sync_batchnorm    train.py

log:

(EAMM) root@ubuntu-desktop:~/EAMM# tree log -L 2
log
├── 124_52000.pth.tar
├── 1-6000.pth.tar
└── 5-3000.pth.tar

What's going on? ./shape_predictor_68_face_landmarks.dat is missing or not submit?

Hello,
I saw you provide the *npy file in test/pose/ folder, the shape is (1, 7).
I changed https://github.com/cleardusk/3DDFA_V2/blob/v0.12/demo.py#L91 default to "pose", the result is like
https://github.com/cleardusk/3DDFA_V2/blob/v0.12/docs/images/trump_hillary_pose.jpg.
No variables shape is (1, 7).

hello，请问多久会开源训练流程呢~可以先提供基于现有数据集的方法么
想进行学习哈 多谢多谢

hi xinya,

i found the AT_net2 use jaco_net = 'cnn', but the model weights you provide is 'gan', so configs and weights are not match, hope you can revise it soon, thx

duomin

I used your pre-trained model, but the results are very strange. Do you happen to know why this is?

Thanks for your contribution!
I wonder where should I download the datasets for training and orginize them? It seems that there is no data in the link you provide.

Hi,
First, thanks for the awesome repo. I am encountering the following error while trying to run the demo.py

Module sync_batchnorm not found
This is happening due to the call in Line-7 of modules/util.py (this line)

Can you kindly suggest a fix for this ?

Thanks,
Avisek

Hello,

Thank you for your contribution. I'm trying to run training, but it seems like we're missing the videos at VoxCeleb1_Cut/right.npy:

FileNotFoundError: [Errno 2] No such file or directory: '/mnt/lustre/share_data/jixinya/VoxCeleb1_Cut/right.npy'   

Can you provide this file?

Hello, I found a few errors in run.py file:

1. You have an extra space in from of train_part2 function on the last line of run.py
2. ImportError: cannot import name 'get_logger' from 'modules.util', so basically there is no get_logger in modules.util. Could you fix it?

你好，请问预训练模型的下载链接在哪？

Hi! Thank you for sharing your code. When I run the test code, the right eye is much larger than the left eye in the result. Is that correct or some config parameters might be wrong? Can I fix this through finetuning it on my own dataset?

Thank you for the great work. I wanted to ask regarding the parameter "--pose_given" that is used in demo.py file, how do I get this for an image?

Hello, I found a few errors in train.py:

1. Traceback (most recent call last):
   File "/fsx/nikitadrobyshev/EAMM/run.py", line 121, in
   train_part1(config, generator, discriminator, kp_detector, kp_detector_a,audio_feature, opt.checkpoint, opt.audio_checkpoint, log_dir, dataset, test_dataset,opt.device_ids, name)
   File "/fsx/nikitadrobyshev/EAMM/train.py", line 26, in train_part1
   optimizer_generator, optimizer_discriminator,
   NameError: name 'optimizer_generator' is not defined

2. File "/fsx/nikitadrobyshev/EAMM/train.py", line 55, in train_part1
   generator_full = TrainPart1Model(kp_detector, kp_detector_a, audio_feature, generator, discriminator, train_params,device_ids)
   File "/fsx/nikitadrobyshev/EAMM/modules/model.py", line 164, in init
   super(TrainFullModel, self).init()
   NameError: name 'TrainFullModel' is not defined

Hello,
I am trying to generate some videos using the pre-trained models, however even though I change the emotion the code seems to generate videos for the same emotion, therefore I was wondering if the Emotion Displacement Learner should be train for each emotion.
Thanks so much,
Jenifer