update
This commit is contained in:
MeeeyoAI
+48
-33
@@ -155,6 +155,10 @@ class LoadAndAdjustImage:
|
||||
|
||||
|
||||
#======重置图像
|
||||
import torch
|
||||
import numpy as np
|
||||
from PIL import Image, ImageOps
|
||||
|
||||
class ImageAdjuster:
|
||||
@classmethod
|
||||
def INPUT_TYPES(s):
|
||||
@@ -165,7 +169,8 @@ class ImageAdjuster:
|
||||
"size_option": ([
|
||||
"Custom", "Million Pixels", "Small", "Medium", "Large",
|
||||
"480P-H(vid 4:3)", "480P-V(vid 3:4)", "720P-H(vid 16:9)", "720P-V(vid 9:16)", "832×480", "480×832"],
|
||||
{"default": "Million Pixels"})
|
||||
{"default": "Million Pixels"}
|
||||
)
|
||||
}
|
||||
}
|
||||
|
||||
@@ -179,18 +184,23 @@ class ImageAdjuster:
|
||||
return float("NaN")
|
||||
|
||||
def process_image(self, image, max_dimension=1024, size_option="Custom"):
|
||||
input_image = Image.fromarray((image.squeeze(0).numpy() * 255).astype(np.uint8))
|
||||
W, H = input_image.size
|
||||
|
||||
batch_size = image.shape[0]
|
||||
processed_images = []
|
||||
widths = []
|
||||
heights = []
|
||||
|
||||
for frame in [input_image]:
|
||||
frame = ImageOps.exif_transpose(frame)
|
||||
for i in range(batch_size):
|
||||
current_image = image[i]
|
||||
input_pil_image = Image.fromarray((current_image.numpy() * 255).astype(np.uint8))
|
||||
W, H = input_pil_image.size
|
||||
|
||||
if frame.mode == 'P':
|
||||
frame = frame.convert("RGBA")
|
||||
elif 'A' in frame.getbands():
|
||||
frame = frame.convert("RGBA")
|
||||
processed_image_pil = input_pil_image.copy()
|
||||
processed_image_pil = ImageOps.exif_transpose(processed_image_pil)
|
||||
|
||||
if processed_image_pil.mode == 'P':
|
||||
processed_image_pil = processed_image_pil.convert("RGBA")
|
||||
elif 'A' in processed_image_pil.getbands():
|
||||
processed_image_pil = processed_image_pil.convert("RGBA")
|
||||
|
||||
if size_option != "Custom":
|
||||
aspect_ratio = W / H
|
||||
@@ -211,50 +221,55 @@ class ImageAdjuster:
|
||||
(1120, 1600) if aspect_ratio <= 0.82 else
|
||||
(1600, 1600)
|
||||
),
|
||||
"Million Pixels": self._resize_to_million_pixels(W, H), # Million Pixels option
|
||||
"480P-H(vid 4:3)": (640, 480), # 480P-H, 640x480
|
||||
"480P-V(vid 3:4)": (480, 640), # 480P-V, 480x640
|
||||
"720P-H(vid 16:9)": (1280, 720), # 720P-H, 1280x720
|
||||
"720P-V(vid 9:16)": (720, 1280), # 720P-V, 720x1280
|
||||
"832×480": (832, 480), # 832x480
|
||||
"480×832": (480, 832), # 480x832
|
||||
"Million Pixels": self._resize_to_million_pixels(W, H),
|
||||
"480P-H(vid 4:3)": (640, 480),
|
||||
"480P-V(vid 3:4)": (480, 640),
|
||||
"720P-H(vid 16:9)": (1280, 720),
|
||||
"720P-V(vid 9:16)": (720, 1280),
|
||||
"832×480": (832, 480),
|
||||
"480×832": (480, 832)
|
||||
}
|
||||
|
||||
target_width, target_height = size_options[size_option]
|
||||
processed_image = frame.convert("RGB")
|
||||
processed_image = ImageOps.fit(processed_image, (target_width, target_height), method=Image.Resampling.BILINEAR, centering=(0.5, 0.5))
|
||||
processed_image_pil = processed_image_pil.convert("RGB")
|
||||
processed_image_pil = ImageOps.fit(processed_image_pil, (target_width, target_height), method=Image.Resampling.BILINEAR, centering=(0.5, 0.5))
|
||||
else:
|
||||
ratio = min(max_dimension / W, max_dimension / H)
|
||||
adjusted_width = round(W * ratio)
|
||||
adjusted_height = round(H * ratio)
|
||||
|
||||
processed_image = frame.convert("RGB")
|
||||
processed_image = processed_image.resize((adjusted_width, adjusted_height), Image.Resampling.BILINEAR)
|
||||
processed_image_pil = processed_image_pil.convert("RGB")
|
||||
processed_image_pil = processed_image_pil.resize((adjusted_width, adjusted_height), Image.Resampling.BILINEAR)
|
||||
|
||||
processed_image = np.array(processed_image).astype(np.float32) / 255.0
|
||||
processed_image = torch.from_numpy(processed_image)[None,]
|
||||
processed_image = np.array(processed_image_pil).astype(np.float32) / 255.0
|
||||
processed_image = torch.from_numpy(processed_image)
|
||||
processed_images.append(processed_image)
|
||||
|
||||
output_image = torch.cat(processed_images, dim=0) if len(processed_images) > 1 else processed_images[0]
|
||||
if size_option != "Custom":
|
||||
return (output_image, target_width, target_height)
|
||||
if size_option != "Custom":
|
||||
widths.append(target_width)
|
||||
heights.append(target_height)
|
||||
else:
|
||||
widths.append(adjusted_width)
|
||||
heights.append(adjusted_height)
|
||||
|
||||
output_image = torch.stack(processed_images)
|
||||
|
||||
if all(w == widths[0] for w in widths) and all(h == heights[0] for h in heights):
|
||||
return (output_image, widths[0], heights[0])
|
||||
else:
|
||||
return (output_image, adjusted_width, adjusted_height)
|
||||
return (output_image, widths[0], heights[0])
|
||||
|
||||
def _resize_to_million_pixels(self, W, H):
|
||||
# Calculate the aspect ratio of the original image
|
||||
aspect_ratio = W / H
|
||||
target_area = 1000000 # 1 million pixels
|
||||
target_area = 1000000
|
||||
|
||||
# Calculate the new width and height while maintaining the aspect ratio
|
||||
if aspect_ratio > 1: # Landscape
|
||||
if aspect_ratio > 1:
|
||||
width = int(np.sqrt(target_area * aspect_ratio))
|
||||
height = int(target_area / width)
|
||||
else: # Portrait
|
||||
else:
|
||||
height = int(np.sqrt(target_area / aspect_ratio))
|
||||
width = int(target_area / height)
|
||||
|
||||
# Round width and height to the nearest multiple of 8
|
||||
width = (width + 7) // 8 * 8
|
||||
height = (height + 7) // 8 * 8
|
||||
|
||||
|
||||
Reference in New Issue
Block a user