Clone this repository. Download the pre-trained Pose-Transfer pytorch model from Pose-Transfer Download the pre-trained Keras model for Human-Pose Detection(Open-pose) from Open-pose. Download the pre-trained VGG-19 PyTorch model from VGG.

Put the Pose-Transfer model in the checkpoints/final_honeywell directory Put the VGG model in the main Data-Augmentation repository. Put the Human-Pose Detection(Open-pose) model in the Data-Augmentation/tool.

python 2
pytorch(0.3.1)
torchvision(0.2.0)
numpy
opencv
scipy
scikit-image
pillow
pandas
tqdm
dominate

place all the input images to transform in /seed_data/test_in

Put all the target images, to which the input image is transformed in seed-data/test_out.

Run the following python commands to generate 18-channel keypoint Posemaps of the input and target images.

python tool/compute_coordinates.py --phase test_in --img_size 128,64 
python tool/compute_coordinates.py --phase test_out --img_size 128,64

Input Posemaps are stored in seed_data/test_inK. Target Posemaps are stored in seed_data/test_outK

The model will take 3 inputs, 1) Input image from test_in, 2) Input pose from test_inK and, 3) Output Pose from test_outK. Every input person image from test_in will be transformed to a set of person images in all the target poses.

python test.py --dataroot ./seed_data/ --name final_honeywell --model PATN --phase test --dataset_mode keypoint --norm batch --batchSize 1 --resize_or_crop no --gpu_ids 0 --BP_input_nc 18 --no_flip --which_model_netG PATN --checkpoints_dir ./checkpoints --which_epoch latest --results_dir ./results --display_id 0