Merge branch 'dev'

.github/workflows/lint.yml

@@ -8,7 +8,6 @@ on:
 jobs:
   lint:
     runs-on: ubuntu-latest
-    if: github.repository == 'Bing-su/adetailer' || github.repository == ''
 
     steps:
       - uses: actions/checkout@v4

.pre-commit-config.yaml

@@ -2,14 +2,18 @@ repos:
   - repo: https://github.com/pre-commit/pre-commit-hooks
     rev: v4.5.0
     hooks:
+      - id: check-added-large-files
+        args: [--maxkb=100]
+      - id: check-merge-conflict
+      - id: check-case-conflict
       - id: check-ast
       - id: trailing-whitespace
         args: [--markdown-linebreak-ext=md]
       - id: end-of-file-fixer
       - id: mixed-line-ending
 
   - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.2.2
+    rev: v0.3.1
     hooks:
       - id: ruff
         args: [--fix, --exit-non-zero-on-fix]

CHANGELOG.md

@@ -1,5 +1,11 @@
 # Changelog
 
+## 2024-03-16
+
+- YOLO World v2, YOLO9 지원가능한 버전으로 ultralytics 업데이트
+- inpaint full res인 경우 인페인트 모드에서 동작하게 변경
+- inpaint full res가 아닌 경우, 사용자가 입력한 마스크와 교차점이 있는 마스크만 선택하여 사용함
+
 ## 2024-03-01
 
 - v24.3.0

README.md

@@ -102,3 +102,11 @@ ADetailer works in three simple steps.
 1. Create an image.
 2. Detect object with a detection model and create a mask image.
 3. Inpaint using the image from 1 and the mask from 2.
+
+## Development
+
+ADetailer is developed and tested using the stable-diffusion 1.5 model, for the [AUTOMATIC1111/stable-diffusion-webui](https://github.com/AUTOMATIC1111/stable-diffusion-webui) repository only.
+
+## License
+
+ADetailer is a derivative work that uses two AGPL-licensed works (stable-diffusion-webui, ultralytics) and is therefore distributed under the AGPL license.

adetailer/__version__.py

@@ -1 +1 @@
-__version__ = "24.3.0"
+__version__ = "24.3.1"

adetailer/args.py

@@ -5,16 +5,28 @@
 from functools import cached_property, partial
 from typing import Any, Literal, NamedTuple, Optional
 
-from pydantic import (
-    BaseModel,
-    Extra,
-    NonNegativeFloat,
-    NonNegativeInt,
-    PositiveInt,
-    confloat,
-    conint,
-    validator,
-)
+try:
+    from pydantic.v1 import (
+        BaseModel,
+        Extra,
+        NonNegativeFloat,
+        NonNegativeInt,
+        PositiveInt,
+        confloat,
+        conint,
+        validator,
+    )
+except ImportError:
+    from pydantic import (
+        BaseModel,
+        Extra,
+        NonNegativeFloat,
+        NonNegativeInt,
+        PositiveInt,
+        confloat,
+        conint,
+        validator,
+    )
 
 
 @dataclass

adetailer/common.py

@@ -3,11 +3,12 @@
 from collections import OrderedDict
 from dataclasses import dataclass, field
 from pathlib import Path
-from typing import Optional
+from typing import Any, Optional
 
 from huggingface_hub import hf_hub_download
 from PIL import Image, ImageDraw
 from rich import print
+from torchvision.transforms.functional import to_pil_image
 
 REPO_ID = "Bingsu/adetailer"
 _download_failed = False
@@ -44,7 +45,7 @@ def scan_model_dir(path_: str | Path) -> list[Path]:
 
 def get_models(
     model_dir: str | Path, extra_dir: str | Path = "", huggingface: bool = True
-) -> OrderedDict[str, str | None]:
+) -> OrderedDict[str, str]:
     model_paths = [*scan_model_dir(model_dir), *scan_model_dir(extra_dir)]
 
     models = OrderedDict()
@@ -56,17 +57,17 @@ def get_models(
                 "hand_yolov8n.pt": hf_download("hand_yolov8n.pt"),
                 "person_yolov8n-seg.pt": hf_download("person_yolov8n-seg.pt"),
                 "person_yolov8s-seg.pt": hf_download("person_yolov8s-seg.pt"),
-                "yolov8x-world.pt": hf_download(
-                    "yolov8x-world.pt", repo_id="Bingsu/yolo-world-mirror"
+                "yolov8x-worldv2.pt": hf_download(
+                    "yolov8x-worldv2.pt", repo_id="Bingsu/yolo-world-mirror"
                 ),
             }
         )
     models.update(
         {
-            "mediapipe_face_full": None,
-            "mediapipe_face_short": None,
-            "mediapipe_face_mesh": None,
-            "mediapipe_face_mesh_eyes_only": None,
+            "mediapipe_face_full": "mediapipe_face_full",
+            "mediapipe_face_short": "mediapipe_face_short",
+            "mediapipe_face_mesh": "mediapipe_face_mesh",
+            "mediapipe_face_mesh_eyes_only": "mediapipe_face_mesh_eyes_only",
         }
     )
 
@@ -133,3 +134,11 @@ def create_bbox_from_mask(
         if bbox is not None:
             bboxes.append(list(bbox))
     return bboxes
+
+
+def ensure_pil_image(image: Any, mode: str = "RGB") -> Image.Image:
+    if not isinstance(image, Image.Image):
+        image = to_pil_image(image)
+    if image.mode != mode:
+        image = image.convert(mode)
+    return image

adetailer/mask.py

@@ -3,13 +3,14 @@
 from enum import IntEnum
 from functools import partial, reduce
 from math import dist
+from typing import Any
 
 import cv2
 import numpy as np
 from PIL import Image, ImageChops
 
 from adetailer.args import MASK_MERGE_INVERT
-from adetailer.common import PredictOutput
+from adetailer.common import PredictOutput, ensure_pil_image
 
 
 class SortBy(IntEnum):
@@ -83,12 +84,19 @@ def offset(img: Image.Image, x: int = 0, y: int = 0) -> Image.Image:
     return ImageChops.offset(img, x, -y)
 
 
-def is_all_black(img: Image.Image) -> bool:
-    arr = np.array(img)
-    return cv2.countNonZero(arr) == 0
+def is_all_black(img: Image.Image | np.ndarray) -> bool:
+    if isinstance(img, Image.Image):
+        img = np.array(img)
+    return cv2.countNonZero(img) == 0
+
+
+def has_intersection(im1: Any, im2: Any) -> bool:
+    arr1 = np.array(ensure_pil_image(im1, "L"))
+    arr2 = np.array(ensure_pil_image(im2, "L"))
+    return not is_all_black(cv2.bitwise_and(arr1, arr2))
 
 
-def bbox_area(bbox: list[float]):
+def bbox_area(bbox: list[float]) -> float:
     return (bbox[2] - bbox[0]) * (bbox[3] - bbox[1])
 
 

controlnet_ext/__init__.py

@@ -13,8 +13,12 @@
         get_cn_models,
     )
 
+from .restore import CNHijackRestore, cn_allow_script_control
+
 __all__ = [
     "ControlNetExt",
+    "CNHijackRestore",
+    "cn_allow_script_control",
     "controlnet_exists",
     "controlnet_type",
     "get_cn_models",

install.py

@@ -44,7 +44,7 @@ def run_pip(*args):
 def install():
     deps = [
         # requirements
-        ("ultralytics", "8.1.18", None),
+        ("ultralytics", "8.1.29", None),
         ("mediapipe", "0.10.9", None),
         ("rich", "13.0.0", None),
         # mediapipe

pyproject.toml

@@ -2,17 +2,41 @@
 name = "adetailer"
 description = "An object detection and auto-mask extension for stable diffusion webui."
 authors = [{ name = "dowon", email = "[email protected]" }]
-requires-python = ">=3.8,<3.12"
+requires-python = ">=3.8,<3.13"
 readme = "README.md"
 license = { text = "AGPL-3.0" }
+dependencies = [
+    "ultralytics>=8.1",
+    "mediapipe>=10",
+    "pydantic<3",
+    "rich>=13",
+    "huggingface_hub",
+]
+keywords = [
+    "stable-diffusion",
+    "stable-diffusion-webui",
+    "adetailer",
+    "ultralytics",
+]
+dynamic = ["version"]
 
 [project.urls]
 repository = "https://github.com/Bing-su/adetailer"
 
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.version]
+path = "adetailer/__version__.py"
+
 [tool.isort]
 profile = "black"
 known_first_party = ["launch", "modules"]
 
+[tool.ruff]
+target-version = "py38"
+
 [tool.ruff.lint]
 select = [
     "A",

scripts/!adetailer.py

@@ -14,9 +14,8 @@
 
 import gradio as gr
 import torch
-from PIL import Image
+from PIL import Image, ImageChops
 from rich import print
-from torchvision.transforms.functional import to_pil_image
 
 import modules
 from adetailer import (
@@ -27,25 +26,25 @@
     ultralytics_predict,
 )
 from adetailer.args import ALL_ARGS, BBOX_SORTBY, ADetailerArgs, SkipImg2ImgOrig
-from adetailer.common import PredictOutput
+from adetailer.common import PredictOutput, ensure_pil_image
 from adetailer.mask import (
     filter_by_ratio,
     filter_k_largest,
+    has_intersection,
+    is_all_black,
     mask_preprocess,
     sort_bboxes,
 )
 from adetailer.traceback import rich_traceback
 from adetailer.ui import WebuiInfo, adui, ordinal, suffix
 from controlnet_ext import (
+    CNHijackRestore,
     ControlNetExt,
+    cn_allow_script_control,
     controlnet_exists,
     controlnet_type,
     get_cn_models,
 )
-from controlnet_ext.restore import (
-    CNHijackRestore,
-    cn_allow_script_control,
-)
 from modules import images, paths, safe, script_callbacks, scripts, shared
 from modules.devices import NansException
 from modules.processing import (
@@ -565,27 +564,24 @@ def sort_bboxes(self, pred: PredictOutput) -> PredictOutput:
         sortby_idx = BBOX_SORTBY.index(sortby)
         return sort_bboxes(pred, sortby_idx)
 
-    def pred_preprocessing(self, pred: PredictOutput, args: ADetailerArgs):
+    def pred_preprocessing(self, p, pred: PredictOutput, args: ADetailerArgs):
         pred = filter_by_ratio(
             pred, low=args.ad_mask_min_ratio, high=args.ad_mask_max_ratio
         )
         pred = filter_k_largest(pred, k=args.ad_mask_k_largest)
         pred = self.sort_bboxes(pred)
-        return mask_preprocess(
+        masks = mask_preprocess(
             pred.masks,
             kernel=args.ad_dilate_erode,
             x_offset=args.ad_x_offset,
             y_offset=args.ad_y_offset,
             merge_invert=args.ad_mask_merge_invert,
         )
-
-    @staticmethod
-    def ensure_rgb_image(image: Any):
-        if not isinstance(image, Image.Image):
-            image = to_pil_image(image)
-        if image.mode != "RGB":
-            image = image.convert("RGB")
-        return image
+        if self.is_img2img_inpaint(p) and not self.is_inpaint_only_masked(p):
+            invert = p.inpainting_mask_invert
+            image_mask = ensure_pil_image(p.image_mask, mode="L")
+            masks = self.inpaint_mask_filter(image_mask, masks, invert)
+        return masks
 
     @staticmethod
     def i2i_prompts_replace(
@@ -637,16 +633,30 @@ def get_each_tap_seed(seed: int, i: int):
 
     @staticmethod
     def is_img2img_inpaint(p) -> bool:
-        return hasattr(p, "image_mask") and bool(p.image_mask)
+        return hasattr(p, "image_mask") and p.image_mask is not None
+
+    @staticmethod
+    def is_inpaint_only_masked(p) -> bool:
+        return hasattr(p, "inpaint_full_res") and p.inpaint_full_res
+
+    @staticmethod
+    def inpaint_mask_filter(
+        img2img_mask: Image.Image, ad_mask: list[Image.Image], invert: int = 0
+    ) -> list[Image.Image]:
+        if invert:
+            img2img_mask = ImageChops.invert(img2img_mask)
+        return [mask for mask in ad_mask if has_intersection(img2img_mask, mask)]
 
     @rich_traceback
     def process(self, p, *args_):
         if getattr(p, "_ad_disabled", False):
             return
 
-        if self.is_img2img_inpaint(p):
+        if self.is_img2img_inpaint(p) and is_all_black(p.image_mask):
             p._ad_disabled = True
-            msg = "[-] ADetailer: img2img inpainting detected. adetailer disabled."
+            msg = (
+                "[-] ADetailer: img2img inpainting with no mask -- adetailer disabled."
+            )
             print(msg)
             return
 
@@ -700,7 +710,7 @@ def _postprocess_image_inner(
         with change_torch_load():
             pred = predictor(ad_model, pp.image, args.ad_confidence, **kwargs)
 
-        masks = self.pred_preprocessing(pred, args)
+        masks = self.pred_preprocessing(p, pred, args)
         shared.state.assign_current_image(pred.preview)
 
         if not masks:
@@ -726,7 +736,7 @@ def _postprocess_image_inner(
         p2 = copy(i2i)
         for j in range(steps):
             p2.image_mask = masks[j]
-            p2.init_images[0] = self.ensure_rgb_image(p2.init_images[0])
+            p2.init_images[0] = ensure_pil_image(p2.init_images[0], "RGB")
             self.i2i_prompts_replace(p2, ad_prompts, ad_negatives, j)
 
             if re.match(r"^\s*\[SKIP\]\s*$", p2.prompt):
@@ -760,7 +770,7 @@ def postprocess_image(self, p, pp, *args_):
             return
 
         pp.image = self.get_i2i_init_image(p, pp)
-        pp.image = self.ensure_rgb_image(pp.image)
+        pp.image = ensure_pil_image(pp.image, "RGB")
         init_image = copy(pp.image)
         arg_list = self.get_args(p, *args_)
         params_txt_content = Path(paths.data_path, "params.txt").read_text("utf-8")