EverybodyDanceNow-Temporal-Smoothing-FaceGAN

Created by Liangjian Chen.

I analysied the video on Windows 10 and train the model on Ubuntu 16.04

Reference:

Part of the project is inherited from:

nyoki-mtl pytorch-EverybodyDanceNow

Lotayou everybody_dance_now_pytorch

OpenPose release　and video pre-processing

Download OpenPose Source and target video:

• Source video can be download from 1.video

• Target video can be download from 2.video

• Download OpenPose release from here

Pre-processing in Windows

Paste 1.mp4 and 2.mp4 under the folder of Open pose release and Run

./build/examples/openpose/openpose.bin --video 1.mp4 --write_json anno_1/ --display 0 --render_pose 0 --face --hand

and

./build/examples/openpose/openpose.bin --video 2.mp4 --write_json anno_2/ --display 0 --render_pose 0 --face --hand

to get the pose annotation from video

Pre-trained models

• Download vgg19-dcbb9e9d.pth.crdownload here and put it in ./src/pix2pixHD/models/
  

• Download pre-trained vgg_16 for face enhancement here and put in ./face_enhancer/

Full process

This step is completed in Ubuntu 16.04

Pose2vid network

Make target pictures

• put the 2.mp4 and anno_2 in ./data/1 and rename it to video.mp4 and anno

• Run python target.py --name 1

Make source pictures

• put the 1.mp4 and anno_1 in ./data/1 and rename it to video.mp4 and anno

• Run python source.py --name 1 --which_train 2

• source.py rescales the label and save it in ./data/2/test/

Train and use pose2vid network

• Run python train_pose2vid_temporal.py and check loss and full training process in ./checkpoints/

• If you break the traning and want to continue last training, set load_pretrain = './checkpoints/target/ in ./src/config/train_opt.py

• Run transfer.py and get results in ./result

Face enhancement network

Train and use face enhancement network

• Run python ./Face_GAN/prepare_Data and check the results in ./Face_GAN/data/.
• Run python ./Face_GAN/train_face_gan.py train face enhancer and run./Face_GAN/Inference.py to gain results
  

Gain results

• Run python transfer_temporal.py and make result pictures to video

TODO

• Pose estimation
  • [x] Pose
  • [x] Face
  • [x] Hand
• [x] pix2pixHD
• [x] FaceGAN
• [X] Temporal smoothing

Environments

Ubuntu 16.04
Python 3.6.5
Pytorch 0.4.1
OpenCV 3.4.4