Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Support batch processing #51

Open
wants to merge 5 commits into
base: main
Choose a base branch
from
Open

Support batch processing #51

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
59ed595
support batch processing
May 11, 2024
7594ff6
working batch support
Dec 26, 2024
b00f86d
Merge branch 'pythongosssss:main' into add-batch-support
jkrauss82 Dec 26, 2024
6650cb2
remove commented code
Dec 26, 2024
bf9813a
move new batch_size parameter to optional parameters to not break leg…
Dec 26, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
106 changes: 68 additions & 38 deletions wd14tagger.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,7 @@ def get_installed_models():
return models


async def tag(image, model_name, threshold=0.35, character_threshold=0.85, exclude_tags="", replace_underscore=True, trailing_comma=False, client_id=None, node=None):
async def tag(batch, model_name, threshold=0.35, character_threshold=0.85, exclude_tags="", replace_underscore=True, trailing_comma=False, client_id=None, node=None):
if model_name.endswith(".onnx"):
model_name = model_name[0:-5]
installed = list(get_installed_models())
Expand All @@ -60,16 +60,17 @@ async def tag(image, model_name, threshold=0.35, character_threshold=0.85, exclu
input = model.get_inputs()[0]
height = input.shape[1]

# Reduce to max size and pad with white
ratio = float(height)/max(image.size)
new_size = tuple([int(x*ratio) for x in image.size])
image = image.resize(new_size, Image.LANCZOS)
square = Image.new("RGB", (height, height), (255, 255, 255))
square.paste(image, ((height-new_size[0])//2, (height-new_size[1])//2))
for i in range(len(batch)):
# Reduce to max size and pad with white
ratio = float(height)/max(batch[i].size)
new_size = tuple([int(x*ratio) for x in batch[i].size])
batch[i] = batch[i].resize(new_size, Image.LANCZOS)
square = Image.new("RGB", (height, height), (255, 255, 255))
square.paste(batch[i], ((height-new_size[0])//2, (height-new_size[1])//2))

image = np.array(square).astype(np.float32)
image = image[:, :, ::-1] # RGB -> BGR
image = np.expand_dims(image, 0)
batch[i] = np.array(square).astype(np.float32)
batch[i] = batch[i][:, :, ::-1] # RGB -> BGR
batch[i] = np.expand_dims(batch[i], 0)

# Read all tags from csv and locate start of each category
tags = []
Expand All @@ -88,22 +89,32 @@ async def tag(image, model_name, threshold=0.35, character_threshold=0.85, exclu
else:
tags.append(row[1])

# imgs = np.array([im for im in batch])

probs = []
label_name = model.get_outputs()[0].name
probs = model.run([label_name], {input.name: image})[0]
for img in batch:
probs.append(model.run([label_name], {input.name: img})[0])
# probs = probs[: len(batch)]
# probs = model.run([label_name], {input.name: imgs})[0]

# print(probs)

result = list(zip(tags, probs[0]))
res = []

# rating = max(result[:general_index], key=lambda x: x[1])
general = [item for item in result[general_index:character_index] if item[1] > threshold]
character = [item for item in result[character_index:] if item[1] > character_threshold]
for i in range(len(batch)):
result = list(zip(tags, probs[i][0]))

all = character + general
remove = [s.strip() for s in exclude_tags.lower().split(",")]
all = [tag for tag in all if tag[0] not in remove]
# rating = max(result[:general_index], key=lambda x: x[1])
general = [item for item in result[general_index:character_index] if item[1] > threshold]
character = [item for item in result[character_index:] if item[1] > character_threshold]

res = ("" if trailing_comma else ", ").join((item[0].replace("(", "\\(").replace(")", "\\)") + (", " if trailing_comma else "") for item in all))
all = character + general
remove = [s.strip() for s in exclude_tags.lower().split(",")]
all = [tag for tag in all if tag[0] not in remove]

res.append(("" if trailing_comma else ", ").join((item[0].replace("(", "\\(").replace(")", "\\)") + (", " if trailing_comma else "") for item in all)))

print(res)
return res


Expand Down Expand Up @@ -172,15 +183,21 @@ class WD14Tagger:
def INPUT_TYPES(s):
extra = [name for name, _ in (os.path.splitext(m) for m in get_installed_models()) if name not in known_models]
models = known_models + extra
return {"required": {
"image": ("IMAGE", ),
"model": (models, { "default": defaults["model"] }),
"threshold": ("FLOAT", {"default": defaults["threshold"], "min": 0.0, "max": 1, "step": 0.05}),
"character_threshold": ("FLOAT", {"default": defaults["character_threshold"], "min": 0.0, "max": 1, "step": 0.05}),
"replace_underscore": ("BOOLEAN", {"default": defaults["replace_underscore"]}),
"trailing_comma": ("BOOLEAN", {"default": defaults["trailing_comma"]}),
"exclude_tags": ("STRING", {"default": defaults["exclude_tags"]}),
}}
return {
"required": {
"image": ("IMAGE", ),
"model": (models, { "default": defaults["model"] }),
"threshold": ("FLOAT", {"default": defaults["threshold"], "min": 0.0, "max": 1, "step": 0.05}),
"character_threshold": ("FLOAT", {"default": defaults["character_threshold"], "min": 0.0, "max": 1, "step": 0.05}),
"replace_underscore": ("BOOLEAN", {"default": defaults["replace_underscore"]}),
"trailing_comma": ("BOOLEAN", {"default": defaults["trailing_comma"]}),
"exclude_tags": ("STRING", {"default": defaults["exclude_tags"]}),

},
"optional": {
"batch_size": ("INT", {"default": 1, "min": 1, "max": 128}),
}
}

RETURN_TYPES = ("STRING",)
OUTPUT_IS_LIST = (True,)
Expand All @@ -189,18 +206,31 @@ def INPUT_TYPES(s):

CATEGORY = "image"

def tag(self, image, model, threshold, character_threshold, exclude_tags="", replace_underscore=False, trailing_comma=False):
tensor = image*255
tensor = np.array(tensor, dtype=np.uint8)
def tag(self, image, model, threshold, character_threshold, exclude_tags="", replace_underscore=False, trailing_comma=False, batch_size=1):
if not isinstance(image, list):
images = [image]
else:
images = image

pbar = comfy.utils.ProgressBar(tensor.shape[0])
pbar = comfy.utils.ProgressBar(len(images))
tags = []
for i in range(tensor.shape[0]):
image = Image.fromarray(tensor[i])
tags.append(wait_for_async(lambda: tag(image, model, threshold, character_threshold, exclude_tags, replace_underscore, trailing_comma)))
pbar.update(1)
return {"ui": {"tags": tags}, "result": (tags,)}
batch = []

for image in images:
tensor = image*255
tensor = np.array(tensor, dtype=np.uint8)

for i in range(tensor.shape[0]):
image = Image.fromarray(tensor[i])
batch.append(image)
if len(batch) == batch_size or i == tensor.shape[0] -1:
tags = tags + wait_for_async(lambda: tag(batch, model, threshold, character_threshold, exclude_tags, replace_underscore, trailing_comma))
pbar.update(len(batch))
batch = []

print(tags)

return {"ui": {"tags": tags}, "result": (tags,)}

NODE_CLASS_MAPPINGS = {
"WD14Tagger|pysssss": WD14Tagger,
Expand Down