ikhwan12 / stylegan

Face Generator

Steps

Image Alignment

$ python align_images.py raw_images/ aligned_images/

Parameters :

• output_size - The dimension of images for input to the model (default=1024)
• x_scale - Scaling factor for x dimension (default=1)
• y_scale - Scaling factor for y dimension (default=1)
• em_scale - Scaling factor for eye-mouth distance (default=0.1)
• use_alpha - Add an alpha channel for masking (default=False)

Latent Vector Encoding

$ python encode_images.py --batch_size=2 --output_video=False aligned_images/ generated_images/ latent_representations/ --model_url https://drive.google.com/uc?id=1oGj5qJcbk4Mt38g1k30Bhr_awZc_beM4

Parameters :

• data_dir - Directory for storing optional models (default='data')
• mask_dir - Directory for storing optional masks (default='masks')
• load_last - Start with embeddings from directory(default='')
• dlatent_avg - Use dlatent from file specified here for truncation instead of dlatent_avg from Gs (default='')
• model_res - The dimension of images in the StyleGAN model (default=1024)
• batch_size - Batch size for generator and perceptual model (default=1)
• optimizer - Optimization algorithm used for optimizing dlatents (default='ggt')
• image_size - Size of images for perceptual model (default=256)
• resnet_image_size - Size of images for the Resnet model (default=256)
• lr - Learning rate for perceptual model (default=0.25)
• decay_rate - Decay rate for learning rate (default=0.9)
• iterations - Number of optimization steps for each batch (default=100)
• decay_steps - Decay steps for learning rate decay (as a percent of iterations) (default=4)
• early_stopping - Stop early once training stabilizes (default=True)
• early_stopping_threshold - Stop after this threshold has been reached (default=0.5)
• early_stopping_patience - Number of iterations to wait below threshold (default=10)
• load_effnet - Model to load for EfficientNet approximation of dlatents (default='data/finetuned_effnet.h5')
• load_resnet - Model to load for ResNet approximation of dlatents (default='data/finetuned_resnet.h5')
• use_preprocess_input - Call process_input() first before using feed forward net (default=True)
• use_best_loss - Output the lowest loss value found as the solution (default=True)
• average_best_loss - Do a running weighted average with the previous best dlatents found (default=0.25)
• sharpen_input - Sharpen the input images (default=True)
• use_vgg_loss - Use VGG perceptual loss; 0 to disable, > 0 to scale. (default=0.4)
• use_vgg_layer - Pick which VGG layer to use (default=9)
• use_pixel_loss - Use logcosh image pixel loss; 0 to disable, > 0 to scale (default=1.5)
• use_mssim_loss - Use MS-SIM perceptual loss; 0 to disable, > 0 to scale (default=200)
• randomize_noise - Add noise to dlatents during optimization (default=False)
• load_mask - Load segmentation masks (default=False)
• face_mask - Generate a mask for predicting only the face area (default=True)

Face Prediction (Average Method)

$ python predict.py

Parameters :

• model_path - StyleGAN Model Path (default='model/model.pkl')
• latent_img1 - Latent representation image 1 path (default='latent_representations/0001_01.npy')
• latent_img2 - Latent representation image 2 path (default='latent_representations/0002_01.npy')
• w1 - Weight for image 1 (default=0.6)
• w2 - Weight for image 2 (default=0.4)
• out - Output path with extension (default='result.png')
• age - Age Coefficient where greater is getting younger (default=0)
  • -2.0 : Elder
  • -1.0 : Adult
  • 0.0 : Teen
  • 1.0 : Kid
  • 2.0 : Toddler
• gender - Gender Coefficient where 0.5 is male and -0.5 is female (default=0)

Face Morpher

$ python morpher.py --images=<images_dir_path> --background=average --out_video=<output_path>

Parameters :

• src - Filepath to source image (.jpg, .jpeg, .png)
• dest - Filepath to destination image (.jpg, .jpeg, .png)
• images - Folderpath to images
• width - Custom width of the images/video [default: 500]
• height - Custom height of the images/video [default: 600]
• num - Number of morph frames [default: 20]
• fps - Number frames per second for the video [default: 10]
• out_frames - Folder path to save all image frames
• out_video - Filename to save a video
• plot - Flag to plot images to result.png [default: False]
• background - background of images to be one of (black|transparent|average) [default: black]
• version - Show version