14 KiB
14 KiB
Documentation
- Class name: ImageAutoCrop
- Category: 😺dzNodes/LayerUtility
- Output node: False
- Repo Ref: https://github.com/chflame163/ComfyUI_LayerStyle
自动抠图并按照遮罩裁切图片。可指定生成图片的背景颜色、长宽比和大小。这个节点是为生成训练模型的图片素材而设计的。 *请参照 SegmentAnythingUltra 和 RemBgUltra 节点的模型安装方法安装模型。
ImageAutoCrop的V2升级版
增加mask可选输入。当有mask输入时，直接使用该输入跳过内置遮罩生成。 增加fill_background, 当此项设置为False时将不处理背景，并且超出画幅的部分不纳入输出范围。 aspect_ratio增加original(原始画面宽高比)选项。 scale_by: 允许按长边、短边、宽度或高度指定尺寸缩放。 scale_by_length: 这里的数值作为scale_by指定边的长度。
Input types
Required
-
image
- 输入图像。
- Comfy dtype: IMAGE
- Python dtype: torch.Tensor
-
background_color
- 背景颜色。
- Comfy dtype: STRING
- Python dtype: str
-
aspect_ratio
- 输出的宽高比。这里提供了常见的画幅比例， "custom"为自定义比例。
- Comfy dtype: STRING
- Python dtype: str
-
proportional_width
- 比例宽。如果aspect_ratio选项不是"custom"，此处设置将被忽略。
- Comfy dtype: INT
- Python dtype: int
-
proportional_height
- 比例高。如果aspect_ratio选项不是"custom"，此处设置将被忽略。
- Comfy dtype: INT
- Python dtype: int
-
scale_to_longest_side
- 允许按长边尺寸缩放。
- Comfy dtype: BOOLEAN
- Python dtype: bool
-
longest_side
- scale_to_longest_side被设置为True时，此项将作为是图像长边的长度。
- Comfy dtype: INT
- Python dtype: int
-
detect
- 探测方法，min_bounding_rect是最小外接矩形, max_inscribed_rect是最大内接矩形。
- Comfy dtype: STRING
- Python dtype: str
-
border_reserve
- 保留边框。在探测到的遮罩主体区域之外扩展裁切范围。
- Comfy dtype: INT
- Python dtype: int
-
ultra_detail_range
- 遮罩边缘超精细处理范围，0为不处理，可以节省生成时间。
- Comfy dtype: INT
- Python dtype: int
-
matting_method
- 生成遮罩的方法。有Segment Anything和 RMBG 1.4两种方法。RMBG 1.4运行速度更快。
- Comfy dtype: STRING
- Python dtype: str
-
sam_model
- 此处选择Segment Anything所使用的sam模型。
- Comfy dtype: STRING
- Python dtype: str
-
grounding_dino_model
- 此处选择Segment Anything所使用的grounding_dino模型。
- Comfy dtype: STRING
- Python dtype: str
-
sam_threshold
- Segment Anything的阈值。
- Comfy dtype: FLOAT
- Python dtype: float
-
sam_prompt
- Segment Anything的提示词。
- Comfy dtype: STRING
- Python dtype: str
-
fill_background
- 当此项设置为False时将不处理背景，并且超出画幅的部分不纳入输出范围。
- Comfy dtype: BOOLEAN
- Python dtype: bool
-
scale_by
- 允许按长边、短边、宽度或高度指定尺寸缩放。
- Comfy dtype: STRING
- Python dtype: str
-
scale_by_length
- 这里的数值作为scale_by指定边的长度。
- Comfy dtype: INT
- Python dtype: int
Optional
- mask
- 输入遮罩。当有mask输入时，直接使用该输入跳过内置遮罩生成。
- Comfy dtype: MASK
- Python dtype: torch.Tensor
Output types
-
cropped_image
- 裁切并更换背景后的图像。
- Comfy dtype: IMAGE
- Python dtype: torch.Tensor
-
box_preview
- 裁切位置预览。
- Comfy dtype: IMAGE
- Python dtype: torch.Tensor
-
cropped_mask
- 裁切后的遮罩。
- Comfy dtype: MASK
- Python dtype: torch.Tensor
Usage tips
- Infra type: CPU
Source code
class ImageAutoCropV2:
def __init__(self):
pass
@classmethod
def INPUT_TYPES(self):
matting_method_list = ['RMBG 1.4', 'SegmentAnything']
detect_mode = ['min_bounding_rect', 'max_inscribed_rect', 'mask_area']
ratio_list = ['1:1', '3:2', '4:3', '16:9', '2:3', '3:4', '9:16', 'custom', 'detect_mask', 'original']
scale_to_side_list = ['None', 'longest', 'shortest', 'width', 'height']
return {
"required": {
"image": ("IMAGE", ), #
"fill_background": ("BOOLEAN", {"default": True}), # 是否填充背景
"background_color": ("STRING", {"default": "#FFFFFF"}), # 背景颜色
"aspect_ratio": (ratio_list,),
"proportional_width": ("INT", {"default": 1, "min": 1, "max": 999, "step": 1}),
"proportional_height": ("INT", {"default": 1, "min": 1, "max": 999, "step": 1}),
"scale_to_side": (scale_to_side_list,), # 是否按长边缩放
"scale_to_length": ("INT", {"default": 1024, "min": 4, "max": 999999, "step": 1}),
"detect": (detect_mode,),
"border_reserve": ("INT", {"default": 100, "min": -9999, "max": 9999, "step": 1}),
"ultra_detail_range": ("INT", {"default": 0, "min": 0, "max": 256, "step": 1}),
"matting_method": (matting_method_list,),
"sam_model": (list_sam_model(),),
"grounding_dino_model": (list_groundingdino_model(),),
"sam_threshold": ("FLOAT", {"default": 0.3, "min": 0, "max": 1.0, "step": 0.01}),
"sam_prompt": ("STRING", {"default": "subject"}),
},
"optional": {
"mask": ("MASK",), #
}
}
RETURN_TYPES = ("IMAGE", "IMAGE", "MASK",)
RETURN_NAMES = ("cropped_image", "box_preview", "cropped_mask",)
FUNCTION = 'image_auto_crop_v2'
CATEGORY = '😺dzNodes/LayerUtility'
def image_auto_crop_v2(self, image, fill_background, background_color, aspect_ratio,
proportional_width, proportional_height,
scale_to_side, scale_to_length, detect, border_reserve,
ultra_detail_range, matting_method,
sam_model, grounding_dino_model, sam_threshold, sam_prompt,
mask=None,
):
ret_images = []
ret_box_previews = []
ret_masks = []
input_images = []
input_masks = []
crop_boxs = []
global SAM_MODEL
global DINO_MODEL
global previous_sam_model
global previous_dino_model
for l in image:
input_images.append(torch.unsqueeze(l, 0))
m = tensor2pil(l)
if m.mode == 'RGBA':
input_masks.append(m.split()[-1])
if mask is not None:
if mask.dim() == 2:
mask = torch.unsqueeze(mask, 0)
input_masks = []
for m in mask:
input_masks.append(tensor2pil(torch.unsqueeze(m, 0)).convert('L'))
if len(input_masks) > 0 and len(input_masks) != len(input_images):
input_masks = []
log(f"Warning, {NODE_NAME} unable align alpha to image, drop it.", message_type='warning')
fit = 'letterbox'
if aspect_ratio == 'custom':
ratio = proportional_width / proportional_height
elif aspect_ratio == 'original':
_image = tensor2pil(input_images[0])
ratio = _image.width / _image.height
elif aspect_ratio == 'detect_mask':
ratio = 0
fit = 'fill'
else:
s = aspect_ratio.split(":")
ratio = int(s[0]) / int(s[1])
for i in range(len(input_images)):
_image = tensor2pil(input_images[i]).convert('RGB')
if len(input_masks) > 0:
_mask = input_masks[i]
else:
if matting_method == 'SegmentAnything':
if previous_sam_model != sam_model:
SAM_MODEL = load_sam_model(sam_model)
previous_sam_model = sam_model
if previous_dino_model != grounding_dino_model:
DINO_MODEL = load_groundingdino_model(grounding_dino_model)
previous_dino_model = grounding_dino_model
item = _image.convert('RGBA')
boxes = groundingdino_predict(DINO_MODEL, item, sam_prompt, sam_threshold)
(_, _mask) = sam_segment(SAM_MODEL, item, boxes)
_mask = mask2image(_mask[0])
else:
_mask = RMBG(_image)
if ultra_detail_range:
_mask = tensor2pil(mask_edge_detail(input_images[i], pil2tensor(_mask), ultra_detail_range, 0.01, 0.99))
bluredmask = gaussian_blur(_mask, 20).convert('L')
x = 0
y = 0
width = 0
height = 0
x_offset = 0
y_offset = 0
if detect == "min_bounding_rect":
(x, y, width, height) = min_bounding_rect(bluredmask)
elif detect == "max_inscribed_rect":
(x, y, width, height) = max_inscribed_rect(bluredmask)
else:
(x, y, width, height) = mask_area(bluredmask)
canvas_width, canvas_height = _image.size
x1 = x - border_reserve
y1 = y - border_reserve
x2 = x + width + border_reserve
y2 = y + height + border_reserve
if x1 < 0:
if fill_background:
canvas_width -= x1
x_offset = -x1
else:
x1 = 0
if y1 < 0:
if fill_background:
canvas_height -= y1
y_offset = -y1
else:
y1 = 0
if x2 > _image.width:
if fill_background:
canvas_width += x2 - _image.width
else:
x2 = _image.width
if y2 > _image.height:
if fill_background:
canvas_height += y2 - _image.height
else:
y2 = _image.height
if fill_background:
crop_box = (x1 + x_offset, y1 + y_offset, width + border_reserve*2, height + border_reserve*2)
else:
crop_box = (x1, y1, x2 - x1, y2 - y1)
crop_boxs.append(crop_box)
if len(crop_boxs) > 0: # 批量图强制使用同一尺寸
crop_box = crop_boxs[0]
orig_width = crop_box[2]
orig_height = crop_box[3]
if aspect_ratio == 'detect_mask':
ratio = orig_width / orig_height
# calculate target width and height
if orig_width > orig_height:
if scale_to_side == 'longest':
target_width = scale_to_length
target_height = int(target_width / ratio)
elif scale_to_side == 'shortest':
target_height = scale_to_length
target_width = int(target_height * ratio)
elif scale_to_side == 'width':
target_width = scale_to_length
target_height = int(target_width / ratio)
elif scale_to_side == 'height':
target_height = scale_to_length
target_width = int(target_height * ratio)
else:
target_width = orig_width
target_height = int(target_width / ratio)
else:
if scale_to_side == 'longest':
target_height = scale_to_length
target_width = int(target_height * ratio)
elif scale_to_side == 'shortest':
target_width = scale_to_length
target_height = int(target_width / ratio)
elif scale_to_side == 'width':
target_width = scale_to_length
target_height = int(target_width / ratio)
elif scale_to_side == 'height':
target_height = scale_to_length
target_width = int(target_height * ratio)
else:
target_height = orig_height
target_width = int(target_height * ratio)
_canvas = Image.new('RGB', size=(canvas_width, canvas_height), color=background_color)
_mask_canvas = Image.new('L', size=(canvas_width, canvas_height), color='black')
if ultra_detail_range:
_image = pixel_spread(_image, _mask)
if fill_background:
_canvas.paste(_image, box=(x_offset, y_offset), mask=_mask.convert('L'))
else:
_canvas.paste(_image, box=(x_offset, y_offset))
_mask_canvas.paste(_mask, box=(x_offset, y_offset))
preview_image = Image.new('RGB', size=(canvas_width, canvas_height), color='gray')
preview_image.paste(_mask, box=(x_offset, y_offset))
preview_image = draw_rect(preview_image,
crop_box[0], crop_box[1], crop_box[2], crop_box[3],
line_color="#F00000", line_width=(canvas_width + canvas_height)//200)
ret_image = _canvas.crop((crop_box[0], crop_box[1], crop_box[0]+crop_box[2], crop_box[1]+crop_box[3]))
ret_image = fit_resize_image(ret_image, target_width, target_height,
fit=fit, resize_sampler=Image.LANCZOS,
background_color=background_color)
ret_mask = _mask_canvas.crop((crop_box[0], crop_box[1], crop_box[0]+crop_box[2], crop_box[1]+crop_box[3]))
ret_mask = fit_resize_image(ret_mask, target_width, target_height,
fit=fit, resize_sampler=Image.LANCZOS,
background_color="#000000")
ret_images.append(pil2tensor(ret_image))
ret_box_previews.append(pil2tensor(preview_image))
ret_masks.append(image2mask(ret_mask))
log(f"{NODE_NAME} Processed {len(ret_images)} image(s).", message_type='finish')
return (torch.cat(ret_images, dim=0),
torch.cat(ret_box_previews, dim=0),
torch.cat(ret_masks, dim=0),
)