This is a pix2pix demo that learns from pose and translates this into a human. A webcam-enabled application is also provided that translates your pose to the trained pose.

# Clone this repo
git clone [email protected]:GordonRen/pose2pose.git

# Create the conda environment from file
conda env create -f environment.yml

https://github.com/FORTH-ModelBasedTracker/PyOpenPose

python generate_train_data.py --file Panama.mp4

Input:

• file is the name of the video file from which you want to create the data set.

Output:

• Two folders original and landmarks will be created.

If you want to download my dataset, here is also the video file that I used and the generated training dataset (1427 images already split into training and validation).

# Clone the repo from Christopher Hesse's pix2pix TensorFlow implementation
git clone https://github.com/affinelayer/pix2pix-tensorflow.git

# Move the original and landmarks folder into the pix2pix-tensorflow folder
mv pose2pose/landmarks pose2pose/original pix2pix-tensorflow/photos_pose

# Go into the pix2pix-tensorflow folder
cd pix2pix-tensorflow/

# Reset to april version
git reset --hard d6f8e4ce00a1fd7a96a72ed17366bfcb207882c7

# Resize original images
python tools/process.py \
  --input_dir photos_pose/original \
  --operation resize \
  --output_dir photos_pose/original_resized
  
# Resize landmark images
python tools/process.py \
  --input_dir photos_pose/landmarks \
  --operation resize \
  --output_dir photos_pose/landmarks_resized
  
# Combine both resized original and landmark images
python tools/process.py \
  --input_dir photos_pose/landmarks_resized \
  --b_dir photos_pose/original_resized \
  --operation combine \
  --output_dir photos_pose/combined
  
# Split into train/val set
python tools/split.py \
  --dir photos_pose/combined
  
# Train the model on the data
python pix2pix.py \
  --mode train \
  --output_dir pose2pose-model \
  --max_epochs 1000 \
  --input_dir photos_pose/combined/train \
  --which_direction AtoB

For more information around training, have a look at Christopher Hesse's pix2pix-tensorflow implementation.

1. First, we need to reduce the trained model so that we can use an image tensor as input:

   python reduce_model.py --model-input pose2pose-model --model-output pose2pose-reduced-model
   

   Input:

   • model-input is the model folder to be imported.
   • model-output is the model (reduced) folder to be exported.

   Output:

   • It returns a reduced model with less weights file size than the original model.

2. Second, we freeze the reduced model to a single file.

   python freeze_model.py --model-folder pose2pose-reduced-model
   

   Input:

   • model-folder is the model folder of the reduced model.

   Output:

   • It returns a frozen model file frozen_model.pb in the model folder.

I have uploaded a pre-trained frozen model here. This model is trained on 1427 images with epoch 1000.

python pose2pose.py --source 0 --show 2 --tf-model pose2pose-reduced-model/frozen_model.pb

Input:

• source is the device index of the camera (default=0).
• show is an option to display: 0 shows the normal input; 1 shows the pose; 2 shows the normal input and pose (default=2).
• tf-model is the frozen model file.

Example:

• TensorFlow 1.0.0
• PyOpenPose

Kudos to Christopher Hesse for his amazing pix2pix TensorFlow implementation and Gene Kogan for his inspirational workshop.
Inspired by Dat Tran.

See LICENSE for details.