test / cli /process.py
bilegentile's picture
Upload folder using huggingface_hub
c19ca42 verified
# pylint: disable=global-statement
import os
import io
import math
import base64
import numpy as np
import mediapipe as mp
from PIL import Image, ImageOps
from pi_heif import register_heif_opener
from skimage.metrics import structural_similarity as ssim
from scipy.stats import beta
import util
import sdapi
import options
face_model = None
body_model = None
segmentation_model = None
all_images = []
all_images_by_type = {}
class Result():
def __init__(self, typ: str, fn: str, tag: str = None, requested: list = []):
self.type = typ
self.input = fn
self.output = ''
self.basename = ''
self.message = ''
self.image = None
self.caption = ''
self.tag = tag
self.tags = []
self.ops = []
self.steps = requested
def detect_blur(image: Image):
# based on <https://github.com/karthik9319/Blur-Detection/>
bw = ImageOps.grayscale(image)
cx, cy = image.size[0] // 2, image.size[1] // 2
fft = np.fft.fft2(bw)
fftShift = np.fft.fftshift(fft)
fftShift[cy - options.process.blur_samplesize: cy + options.process.blur_samplesize, cx - options.process.blur_samplesize: cx + options.process.blur_samplesize] = 0
fftShift = np.fft.ifftshift(fftShift)
recon = np.fft.ifft2(fftShift)
magnitude = np.log(np.abs(recon))
mean = round(np.mean(magnitude), 2)
return mean
def detect_dynamicrange(image: Image):
# based on <https://towardsdatascience.com/measuring-enhancing-image-quality-attributes-234b0f250e10>
data = np.asarray(image)
image = np.float32(data)
RGB = [0.299, 0.587, 0.114]
height, width = image.shape[:2] # pylint: disable=unsubscriptable-object
brightness_image = np.sqrt(image[..., 0] ** 2 * RGB[0] + image[..., 1] ** 2 * RGB[1] + image[..., 2] ** 2 * RGB[2]) # pylint: disable=unsubscriptable-object
hist, _ = np.histogram(brightness_image, bins=256, range=(0, 255))
img_brightness_pmf = hist / (height * width)
dist = beta(2, 2)
ys = dist.pdf(np.linspace(0, 1, 256))
ref_pmf = ys / np.sum(ys)
dot_product = np.dot(ref_pmf, img_brightness_pmf)
squared_dist_a = np.sum(ref_pmf ** 2)
squared_dist_b = np.sum(img_brightness_pmf ** 2)
res = dot_product / math.sqrt(squared_dist_a * squared_dist_b)
return round(res, 2)
def detect_simmilar(image: Image):
img = image.resize((options.process.similarity_size, options.process.similarity_size))
img = ImageOps.grayscale(img)
data = np.array(img)
similarity = 0
for i in all_images:
val = ssim(data, i, data_range=255, channel_axis=None, gradient=False, full=False)
if val > similarity:
similarity = val
all_images.append(data)
return similarity
def segmentation(res: Result):
global segmentation_model
if segmentation_model is None:
segmentation_model = mp.solutions.selfie_segmentation.SelfieSegmentation(model_selection=options.process.segmentation_model)
data = np.array(res.image)
results = segmentation_model.process(data)
condition = np.stack((results.segmentation_mask,) * 3, axis=-1) > 0.1
background = np.zeros(data.shape, dtype=np.uint8)
background[:] = options.process.segmentation_background
data = np.where(condition, data, background) # consider using a joint bilateral filter instead of pure combine
segmented = Image.fromarray(data)
res.image = segmented
res.ops.append('segmentation')
return res
def unload():
global face_model
if face_model is not None:
face_model = None
global body_model
if body_model is not None:
body_model = None
global segmentation_model
if segmentation_model is not None:
segmentation_model = None
def encode(img):
with io.BytesIO() as stream:
img.save(stream, 'JPEG')
values = stream.getvalue()
encoded = base64.b64encode(values).decode()
return encoded
def reset():
unload()
global all_images_by_type
all_images_by_type = {}
global all_images
all_images = []
def upscale_restore_image(res: Result, upscale: bool = False, restore: bool = False):
kwargs = util.Map({
'image': encode(res.image),
'codeformer_visibility': 0.0,
'codeformer_weight': 0.0,
})
if res.image.width >= options.process.target_size and res.image.height >= options.process.target_size:
upscale = False
if upscale:
kwargs.upscaler_1 = 'SwinIR_4x'
kwargs.upscaling_resize = 2
res.ops.append('upscale')
if restore:
kwargs.codeformer_visibility = 1.0
kwargs.codeformer_weight = 0.2
res.ops.append('restore')
if upscale or restore:
result = sdapi.postsync('/sdapi/v1/extra-single-image', kwargs)
if 'image' not in result:
res.message = 'failed to upscale/restore image'
else:
res.image = Image.open(io.BytesIO(base64.b64decode(result['image'])))
return res
def interrogate_image(res: Result, tag: str = None):
caption = ''
tags = []
for model in options.process.interrogate_model:
json = util.Map({ 'image': encode(res.image), 'model': model })
result = sdapi.postsync('/sdapi/v1/interrogate', json)
if model == 'clip':
caption = result.caption if 'caption' in result else ''
caption = caption.split(',')[0].replace(' a ', ' ').strip()
if tag is not None:
caption = res.tag + ', ' + caption
if model == 'deepdanbooru':
tag = result.caption if 'caption' in result else ''
tags = tag.split(',')
tags = [t.replace('(', '').replace(')', '').replace('\\', '').split(':')[0].strip() for t in tags]
if tag is not None:
for t in res.tag.split(',')[::-1]:
tags.insert(0, t.strip())
pos = 0 if len(tags) == 0 else 1
tags.insert(pos, caption.split(' ')[1])
tags = [t for t in tags if len(t) > 2]
if len(tags) > options.process.tag_limit:
tags = tags[:options.process.tag_limit]
res.caption = caption
res.tags = tags
res.ops.append('interrogate')
return res
def resize_image(res: Result):
resized = res.image
resized.thumbnail((options.process.target_size, options.process.target_size), Image.Resampling.HAMMING)
res.image = resized
res.ops.append('resize')
return res
def square_image(res: Result):
size = max(res.image.width, res.image.height)
squared = Image.new('RGB', (size, size))
squared.paste(res.image, ((size - res.image.width) // 2, (size - res.image.height) // 2))
res.image = squared
res.ops.append('square')
return res
def process_face(res: Result):
res.ops.append('face')
global face_model
if face_model is None:
face_model = mp.solutions.face_detection.FaceDetection(min_detection_confidence=options.process.face_score, model_selection=options.process.face_model)
results = face_model.process(np.array(res.image))
if results.detections is None:
res.message = 'no face detected'
res.image = None
return res
box = results.detections[0].location_data.relative_bounding_box
if box.xmin < 0 or box.ymin < 0 or (box.width - box.xmin) > 1 or (box.height - box.ymin) > 1:
res.message = 'face out of frame'
res.image = None
return res
x = max(0, (box.xmin - options.process.face_pad / 2) * res.image.width)
y = max(0, (box.ymin - options.process.face_pad / 2)* res.image.height)
w = min(res.image.width, (box.width + options.process.face_pad) * res.image.width)
h = min(res.image.height, (box.height + options.process.face_pad) * res.image.height)
x = max(0, x)
res.image = res.image.crop((x, y, x + w, y + h))
return res
def process_body(res: Result):
res.ops.append('body')
global body_model
if body_model is None:
body_model = mp.solutions.pose.Pose(static_image_mode=True, min_detection_confidence=options.process.body_score, model_complexity=options.process.body_model)
results = body_model.process(np.array(res.image))
if results.pose_landmarks is None:
res.message = 'no body detected'
res.image = None
return res
x0 = [res.image.width * (i.x - options.process.body_pad / 2) for i in results.pose_landmarks.landmark if i.visibility > options.process.body_visibility]
y0 = [res.image.height * (i.y - options.process.body_pad / 2) for i in results.pose_landmarks.landmark if i.visibility > options.process.body_visibility]
x1 = [res.image.width * (i.x + options.process.body_pad / 2) for i in results.pose_landmarks.landmark if i.visibility > options.process.body_visibility]
y1 = [res.image.height * (i.y + options.process.body_pad / 2) for i in results.pose_landmarks.landmark if i.visibility > options.process.body_visibility]
if len(x0) < options.process.body_parts:
res.message = f'insufficient body parts detected: {len(x0)}'
res.image = None
return res
res.image = res.image.crop((max(0, min(x0)), max(0, min(y0)), min(res.image.width, max(x1)), min(res.image.height, max(y1))))
return res
def process_original(res: Result):
res.ops.append('original')
return res
def save_image(res: Result, folder: str):
if res.image is None or folder is None:
return res
all_images_by_type[res.type] = all_images_by_type.get(res.type, 0) + 1
res.basename = os.path.basename(res.input).split('.')[0]
res.basename = str(all_images_by_type[res.type]).rjust(3, '0') + '-' + res.type + '-' + res.basename
res.basename = os.path.join(folder, res.basename)
res.output = res.basename + options.process.format
res.image.save(res.output)
res.image.close()
res.ops.append('save')
return res
def file(filename: str, folder: str, tag = None, requested = []):
# initialize result dict
res = Result(fn = filename, typ='unknown', tag=tag, requested = requested)
# open image
try:
register_heif_opener()
res.image = Image.open(filename)
if res.image.mode == 'RGBA':
res.image = res.image.convert('RGB')
res.image = ImageOps.exif_transpose(res.image) # rotate image according to EXIF orientation
except Exception as e:
res.message = f'error opening: {e}'
return res
# primary steps
if 'face' in requested:
res.type = 'face'
res = process_face(res)
elif 'body' in requested:
res.type = 'body'
res = process_body(res)
elif 'original' in requested:
res.type = 'original'
res = process_original(res)
# validation steps
if res.image is None:
return res
if 'blur' in requested:
res.ops.append('blur')
val = detect_blur(res.image)
if val > options.process.blur_score:
res.message = f'blur check failed: {val}'
res.image = None
if 'range' in requested:
res.ops.append('range')
val = detect_dynamicrange(res.image)
if val < options.process.range_score:
res.message = f'dynamic range check failed: {val}'
res.image = None
if 'similarity' in requested:
res.ops.append('similarity')
val = detect_simmilar(res.image)
if val > options.process.similarity_score:
res.message = f'dynamic range check failed: {val}'
res.image = None
if res.image is None:
return res
# post processing steps
res = upscale_restore_image(res, 'upscale' in requested, 'restore' in requested)
if res.image.width < options.process.target_size or res.image.height < options.process.target_size:
res.message = f'low resolution: [{res.image.width}, {res.image.height}]'
res.image = None
return res
if 'interrogate' in requested:
res = interrogate_image(res, tag)
if 'resize' in requested:
res = resize_image(res)
if 'square' in requested:
res = square_image(res)
if 'segment' in requested:
res = segmentation(res)
# finally save image
res = save_image(res, folder)
return res