test_api.py

import os
import tempfile
import torch
import time
import numpy as np
import random
from pathlib import Path
from AdaIN import AdaINNet
from PIL import Image
from torchvision.utils import save_image
from torchvision.transforms import ToPILImage
from utils import adaptive_instance_normalization, grid_image, transform,linear_histogram_matching, Range
from glob import glob
from datasets import load_dataset

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

def style_transfer(content_tensor, style_tensor, encoder, decoder, alpha=1.0):
	"""
	Given content image and style image, generate feature maps with encoder, apply 
	neural style transfer with adaptive instance normalization, generate output image
	with decoder

	Args:
		content_tensor (torch.FloatTensor): Content image 
		style_tensor (torch.FloatTensor): Style Image
		encoder: Encoder (vgg19) network
		decoder: Decoder network
		alpha (float, default=1.0): Weight of style image feature 
	
	Return:
		output_tensor (torch.FloatTensor): Style Transfer output image
	"""

	content_enc = encoder(content_tensor)
	style_enc = encoder(style_tensor)

	transfer_enc = adaptive_instance_normalization(content_enc, style_enc)
	
	mix_enc = alpha * transfer_enc + (1-alpha) * content_enc
	return decoder(mix_enc)

def run_adain(content_dir, style_dataset_pth, out_dir, alpha=1.0, dataset_size=100, vgg_pth='vgg_normalized.pth', decoder_pth='decoder.pth'):
	content_pths = [Path(f) for f in glob(content_dir+'/*')]
	num_content_imgs = len(content_pths)

	assert num_content_imgs > 0, 'Failed to load content image'

	# Load AdaIN model
	vgg = torch.load(vgg_pth)
	model = AdaINNet(vgg).to(device)
	model.decoder.load_state_dict(torch.load(decoder_pth))
	model.eval()
	
	# Prepare image transform
	t = transform(512)
	
	# Timer
	times = []

	style_ds = load_dataset(style_dataset_pth, split="train")

	if num_content_imgs * len(style_ds) > dataset_size:
		num_style_per_content = int(np.ceil(dataset_size / num_content_imgs))
	else:
		num_style_per_content = len(style_ds)

	for content_pth in content_pths:
		content_img = Image.open(content_pth)
		content_tensor = t(content_img).unsqueeze(0).to(device)
		indices = random.sample(range(len(style_ds)), num_style_per_content)

		for idx in indices:
			style_img = style_ds[idx]['image']
			if style_img.mode not in ("RGB", "L"):
				style_img = style_img.convert("RGB")
			style_tensor = t(style_img).unsqueeze(0).to(device)
    
			# Start time
			tic = time.perf_counter()
            
			# Execute style transfer
			with torch.no_grad():
				out_tensor = style_transfer(content_tensor, style_tensor, model.encoder, model.decoder, alpha).cpu()
        
			# End time
			toc = time.perf_counter()
			print("Content: " + content_pth.stem + ". Style: " \
				+ str(style_idx) + '. Alpha: ' + str(alpha) + '. Style Transfer time: %.4f seconds' % (toc-tic))
			times.append(toc-tic)
            
			# Save image
			out_pth = out_dir + content_pth.stem + '_style_' + str(idx) + '_alpha' + str(alpha)
			out_pth += content_pth.suffix
			save_image(out_tensor, out_pth)

	# Remove runtime of first iteration because it is flawed for some unknown reason
	if len(times) > 1:
		times.pop(0)
		avg = sum(times)/len(times)
		print("Average style transfer time: %.4f seconds" % (avg))