def upscale(args, selected_experiment_folder):
target_image = Image.open(
os.path.join(selected_experiment_folder, "target.png"))
upscaling_factor = args.upscaling_factor
original_emoji_size = (args.emoji_size, args.emoji_size)
original_image_size = target_image.size
upscaled_emoji_size = (
original_emoji_size[0] * upscaling_factor,
original_emoji_size[1] * upscaling_factor,
)
upscaled_image_size = (
original_image_size[0] * upscaling_factor,
original_image_size[1] * upscaling_factor,
)
upscaled_emojies = get_emojies(size=upscaled_emoji_size[0])
stored_individual_paths = get_file_paths(selected_experiment_folder,
file_extensions=("pkl", ))
stored_individual_paths.sort(key=lambda f: f.name)
best_stored_individual_path = stored_individual_paths[-1]
genotype = joblib.load(best_stored_individual_path)
genotype[:, 1] *= upscaling_factor # x
genotype[:, 2] *= upscaling_factor # y
image = generate_alpha_image_from_scratch(genotype, upscaled_image_size,
upscaled_emojies)
image.save(
os.path.join(selected_experiment_folder,
best_stored_individual_path.stem + "_upscaled.png"))
print("Done")
os.makedirs(OUTPUT_DIR / experiment_id, exist_ok=True)
target_image.save(os.path.join(OUTPUT_DIR / experiment_id, "target.png"))
target_image = target_image.convert("RGB")
if args.starting_canvas is not None:
starting_canvas = Image.open(args.starting_canvas).convert("RGB")
if should_resize:
starting_canvas = starting_canvas.resize((args.width, args.height),
Image.LANCZOS)
assert starting_canvas.size == target_image.size
Individual.set_starting_canvas(starting_canvas)
Individual.target_image = target_image
Individual.emojies = get_emojies(args.emoji_size)
Individual.max_num_emojies = args.num_generations
print("Found {} emoji images".format(len(Individual.emojies)))
fitness_evaluator_class = FITNESS_EVALUATORS[args.fitness]
fitness_evaluator = fitness_evaluator_class(target_image)
population = [
Individual.get_random_individual() for _ in range(args.population_size)
]
last_saved_fitness = float("-inf")
for i in tqdm(range(args.num_generations)):
fitness_evaluator.evaluate_fitness(population)
ordered_individuals = sorted(population, key=lambda i: i.fitness)
)
arg_parser.add_argument(
'--emoji-size',
dest='emoji_size',
type=positive_int,
required=False,
default=16
)
args = arg_parser.parse_args()
experiment_id = "{}_{}".format(
arrow.utcnow().format("YYYY-MM-DDTHHmm"), uuid.uuid4()
)
target_image = Image.open(TARGET_IMAGES_DIR / args.target).convert("RGB")
emojies = get_emojies(args.emoji_size)
print("Found {} emoji images".format(len(emojies)))
fitness_evaluator_class = FITNESS_EVALUATORS[args.fitness]
fitness_evaluator = fitness_evaluator_class(target_image)
os.makedirs(OUTPUT_DIR / experiment_id, exist_ok=True)
population = [Individual.get_random_individual() for _ in range(population_size)]
last_saved_fitness = float("-inf")
for i in tqdm(range(args.num_generations)):
fitness_evaluator.evaluate_fitness(population)
ordered_individuals = sorted(population, key=lambda i: i.fitness)
fittest_individual = ordered_individuals[-1]
upscaled_image_size = (
original_image_size[0] * upscaling_factor,
original_image_size[1] * upscaling_factor,
)
upscaled_emoji_size = (
original_emoji_size[0] * upscaling_factor,
original_emoji_size[1] * upscaling_factor,
)
stored_individual_paths = get_file_paths(selected_experiment_folder,
file_extensions=("pkl", ))
stored_individual_paths.sort(key=lambda f: f.name)
best_stored_individual_path = stored_individual_paths[-1]
genotype = joblib.load(best_stored_individual_path)
upscaled_emojies = get_emojies(size=upscaled_emoji_size[0])
fitness_evaluator = LABDeltaESSIMFitnessEvaluator(target_image)
image = generate_image_from_scratch(genotype, upscaled_image_size,
upscaled_emojies).convert("RGB")
base_individual = Individual(image)
fitness_evaluator.evaluate_fitness([base_individual])
current_fitness = base_individual.fitness
print("Base fitness: {}".format(current_fitness))
current_genotype = genotype
num_emojies_removed = 0
for i in tqdm(range(len(genotype))):
candidate_genotype = np.copy(current_genotype)
candidate_genotype[i, :] = 0 # Remove this emoji
image = generate_image_from_scratch(candidate_genotype,