import os import torch import gradio as gr import os import sys import numpy as np from e4e.models.psp import pSp from util import * from huggingface_hub import hf_hub_download import os import sys import tempfile import shutil from argparse import Namespace from pathlib import Path import shutil import dlib import numpy as np import torchvision.transforms as transforms from torchvision import utils from PIL import Image from model.sg2_model import Generator from generate_videos import generate_frames, video_from_interpolations, vid_to_gif model_dir = "models" os.makedirs(model_dir, exist_ok=True) model_repos = {"e4e": ("akhaliq/JoJoGAN_e4e_ffhq_encode", "e4e_ffhq_encode.pt"), "dlib": ("akhaliq/jojogan_dlib", "shape_predictor_68_face_landmarks.dat"), "base": ("akhaliq/jojogan-stylegan2-ffhq-config-f", "stylegan2-ffhq-config-f.pt"), "anime": ("rinong/stylegan-nada-models", "anime.pt"), "joker": ("rinong/stylegan-nada-models", "joker.pt") } def get_models(): os.makedirs(model_dir, exist_ok=True) model_paths = {} for model_name, repo_details in model_repos.items(): download_path = hf_hub_download(repo_id=repo_details[0], filename=repo_details[1]) model_paths[model_name] = download_path return model_paths model_paths = get_models() class ImageEditor(object): def __init__(self): self.device = "cuda" if torch.cuda.is_available() else "cpu" latent_size = 512 n_mlp = 8 channel_mult = 2 model_size = 1024 self.generators = {} self.model_list = [name for name in model_paths.keys() if name not in ["e4e", "dlib"]] for model in self.model_list: g_ema = Generator( model_size, latent_size, n_mlp, channel_multiplier=channel_mult ).to(self.device) checkpoint = torch.load(model_paths[model]) g_ema.load_state_dict(checkpoint['g_ema']) self.generators[model] = g_ema self.experiment_args = {"model_path": model_paths["e4e"]} self.experiment_args["transform"] = transforms.Compose( [ transforms.Resize((256, 256)), transforms.ToTensor(), transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5]), ] ) self.resize_dims = (256, 256) model_path = self.experiment_args["model_path"] ckpt = torch.load(model_path, map_location="cpu") opts = ckpt["opts"] opts["checkpoint_path"] = model_path opts = Namespace(**opts) self.e4e_net = pSp(opts, self.device) self.e4e_net.eval() self.shape_predictor = dlib.shape_predictor( model_paths["dlib"] ) print("setup complete") def get_style_list(self): style_list = ['all', 'list - enter below'] for key in self.generators: style_list.append(key) return style_list def predict( self, input, # Input image path output_style, # Which output style do you want to use? style_list, # Comma seperated list of models to use. Only accepts models from the output_style list generate_video, # Generate a video instead of an output image with_editing, # Apply latent space editing to the generated video video_format # Choose gif to display in browser, mp4 for higher-quality downloadable video ): if output_style == 'all': styles = self.model_list elif output_style == 'list - enter below': styles = style_list.split(",") for style in styles: if style not in self.model_list: raise ValueError(f"Encountered style '{style}' in the style_list which is not an available option.") else: styles = [output_style] # @title Align image input_image = self.run_alignment(str(input)) input_image = input_image.resize(self.resize_dims) img_transforms = self.experiment_args["transform"] transformed_image = img_transforms(input_image) with torch.no_grad(): images, latents = self.run_on_batch(transformed_image.unsqueeze(0)) result_image, latent = images[0], latents[0] inverted_latent = latent.unsqueeze(0).unsqueeze(1) out_dir = Path(tempfile.mkdtemp()) out_path = out_dir / "out.jpg" generators = [self.generators[style] for style in styles] if not generate_video: with torch.no_grad(): img_list = [] for g_ema in generators: img, _ = g_ema(inverted_latent, input_is_latent=True, truncation=1, randomize_noise=False) img_list.append(img) out_img = torch.cat(img_list, axis=0) utils.save_image(out_img, out_path, nrow=int(np.sqrt(out_img.size(0))), normalize=True, scale_each=True, range=(-1, 1)) return Path(out_path) return self.generate_vid(generators, inverted_latent, out_dir, video_format, with_editing) def generate_vid(self, generators, latent, out_dir, video_format, with_editing): np_latent = latent.squeeze(0).cpu().detach().numpy() args = { 'fps': 24, 'target_latents': None, 'edit_directions': None, 'unedited_frames': 0 if with_editing else 40 * (len(generators) - 1) } args = Namespace(**args) with tempfile.TemporaryDirectory() as dirpath: generate_frames(args, np_latent, generators, dirpath) video_from_interpolations(args.fps, dirpath) gen_path = Path(dirpath) / "out.mp4" out_path = out_dir / f"out.{video_format}" if video_format == 'gif': vid_to_gif(gen_path, out_dir, scale=256, fps=args.fps) else: shutil.copy2(gen_path, out_path) return out_path def run_alignment(self, image_path): aligned_image = align_face(filepath=image_path, predictor=self.shape_predictor) print("Aligned image has shape: {}".format(aligned_image.size)) return aligned_image def run_on_batch(self, inputs): images, latents = self.e4e_net( inputs.to(self.device).float(), randomize_noise=False, return_latents=True ) return images, latents editor = ImageEditor() title = "StyleGAN-NADA" description = "Gradio Demo for StyleGAN-NADA: CLIP-Guided Domain Adaptation of Image Generators (SIGGRAPH 2022). To use it, upload your image and select a target style. More information about the paper and training new models can be found below." article = "
StyleGAN-NADA: CLIP-Guided Domain Adaptation of Image Generators | Project Page | Code