let5see/IOPaint
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add paint by example

Browse Source
This commit is contained in:
Qing
2022-12-10 22:06:15 +08:00
parent 6e9d3d8442
commit 203f2bc9c7
18 changed files with 572 additions and 82 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
lama_cleaner/model/base.py
View File

@@ -211,6 +211,26 @@ class InpaintModel:
return result
def _apply_cropper(self, image, mask, config: Config):
img_h, img_w = image.shape[:2]
l, t, w, h = (
config.croper_x,
config.croper_y,
config.croper_width,
config.croper_height,
)
r = l + w
b = t + h
l = max(l, 0)
r = min(r, img_w)
t = max(t, 0)
b = min(b, img_h)
crop_img = image[t:b, l:r, :]
crop_mask = mask[t:b, l:r]
return crop_img, crop_mask, (l, t, r, b)
def _run_box(self, image, mask, box, config: Config):
"""

80
lama_cleaner/model/paint_by_example.py
View File

@@ -0,0 +1,80 @@
import random
import PIL
import PIL.Image
import cv2
import numpy as np
import torch
from diffusers import DiffusionPipeline
from lama_cleaner.model.base import InpaintModel
from lama_cleaner.schema import Config
class PaintByExample(InpaintModel):
pad_mod = 8
min_size = 512
def init_model(self, device: torch.device, **kwargs):
use_gpu = device == torch.device('cuda') and torch.cuda.is_available()
torch_dtype = torch.float16 if use_gpu else torch.float32
self.model = DiffusionPipeline.from_pretrained(
"Fantasy-Studio/Paint-by-Example",
torch_dtype=torch_dtype,
)
self.model.enable_attention_slicing()
self.model = self.model.to(device)
def forward(self, image, mask, config: Config):
"""Input image and output image have same size
image: [H, W, C] RGB
mask: [H, W, 1] 255 means area to repaint
return: BGR IMAGE
"""
seed = config.paint_by_example_seed
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
output = self.model(
image=PIL.Image.fromarray(image),
mask_image=PIL.Image.fromarray(mask[:, :, -1], mode="L"),
example_image=config.paint_by_example_example_image,
num_inference_steps=config.paint_by_example_steps,
output_type='np.array',
).images[0]
output = (output * 255).round().astype("uint8")
output = cv2.cvtColor(output, cv2.COLOR_RGB2BGR)
return output
@torch.no_grad()
def __call__(self, image, mask, config: Config):
"""
images: [H, W, C] RGB, not normalized
masks: [H, W]
return: BGR IMAGE
"""
if config.use_croper:
crop_img, crop_mask, (l, t, r, b) = self._apply_cropper(image, mask, config)
crop_image = self._pad_forward(crop_img, crop_mask, config)
inpaint_result = image[:, :, ::-1]
inpaint_result[t:b, l:r, :] = crop_image
else:
inpaint_result = self._pad_forward(image, mask, config)
return inpaint_result
def forward_post_process(self, result, image, mask, config):
if config.paint_by_example_match_histograms:
result = self._match_histograms(result, image[:, :, ::-1], mask)
if config.paint_by_example_mask_blur != 0:
k = 2 * config.paint_by_example_mask_blur + 1
mask = cv2.GaussianBlur(mask, (k, k), 0)
return result, image, mask
@staticmethod
def is_downloaded() -> bool:
# model will be downloaded when app start, and can't switch in frontend settings
return True

59
lama_cleaner/model/sd.py
View File

@@ -12,31 +12,6 @@ from lama_cleaner.model.base import InpaintModel
from lama_cleaner.schema import Config, SDSampler
#
#
# def preprocess_image(image):
# w, h = image.size
# w, h = map(lambda x: x - x % 32, (w, h)) # resize to integer multiple of 32
# image = image.resize((w, h), resample=PIL.Image.LANCZOS)
# image = np.array(image).astype(np.float32) / 255.0
# image = image[None].transpose(0, 3, 1, 2)
# image = torch.from_numpy(image)
# # [-1, 1]
# return 2.0 * image - 1.0
#
#
# def preprocess_mask(mask):
# mask = mask.convert("L")
# w, h = mask.size
# w, h = map(lambda x: x - x % 32, (w, h)) # resize to integer multiple of 32
# mask = mask.resize((w // 8, h // 8), resample=PIL.Image.NEAREST)
# mask = np.array(mask).astype(np.float32) / 255.0
# mask = np.tile(mask, (4, 1, 1))
# mask = mask[None].transpose(0, 1, 2, 3) # what does this step do?
# mask = 1 - mask # repaint white, keep black
# mask = torch.from_numpy(mask)
# return mask
class CPUTextEncoderWrapper:
def __init__(self, text_encoder, torch_dtype):
self.config = text_encoder.config
@@ -92,17 +67,6 @@ class SD(InpaintModel):
return: BGR IMAGE
"""
# image = norm_img(image) # [0, 1]
# image = image * 2 - 1 # [0, 1] -> [-1, 1]
# resize to latent feature map size
# h, w = mask.shape[:2]
# mask = cv2.resize(mask, (h // 8, w // 8), interpolation=cv2.INTER_AREA)
# mask = norm_img(mask)
#
# image = torch.from_numpy(image).unsqueeze(0).to(self.device)
# mask = torch.from_numpy(mask).unsqueeze(0).to(self.device)
scheduler_config = self.model.scheduler.config
if config.sd_sampler == SDSampler.ddim:
@@ -139,7 +103,6 @@ class SD(InpaintModel):
prompt=config.prompt,
negative_prompt=config.negative_prompt,
mask_image=PIL.Image.fromarray(mask[:, :, -1], mode="L"),
strength=config.sd_strength,
num_inference_steps=config.sd_steps,
guidance_scale=config.sd_guidance_scale,
output_type="np.array",
@@ -159,30 +122,10 @@ class SD(InpaintModel):
masks: [H, W]
return: BGR IMAGE
"""
img_h, img_w = image.shape[:2]
# boxes = boxes_from_mask(mask)
if config.use_croper:
logger.info("use croper")
l, t, w, h = (
config.croper_x,
config.croper_y,
config.croper_width,
config.croper_height,
)
r = l + w
b = t + h
l = max(l, 0)
r = min(r, img_w)
t = max(t, 0)
b = min(b, img_h)
crop_img = image[t:b, l:r, :]
crop_mask = mask[t:b, l:r]
crop_img, crop_mask, (l, t, r, b) = self._apply_cropper(image, mask, config)
crop_image = self._pad_forward(crop_img, crop_mask, config)
inpaint_result = image[:, :, ::-1]
inpaint_result[t:b, l:r, :] = crop_image
else: