├── .devcontainer ├── devcontainer.json └── start.sh ├── .github └── workflows │ └── build-image.yml ├── .gitignore ├── .vscode ├── extensions.json └── settings.json ├── Dockerfile ├── LICENSE ├── README.md ├── README_EN.md ├── README_JP.md ├── README_KO.md ├── app.py ├── app.spec ├── assets ├── comfyui.png ├── community-uniapp-wechat-miniprogram.png ├── community-web.png ├── community-wechat-miniprogram.png ├── demo.png ├── demoImage.jpg ├── face++.png ├── gradio-image.jpeg ├── harry.png ├── hivision_logo.png └── social_template.png ├── demo ├── assets │ ├── american-style.png │ ├── color_list_CN.csv │ ├── color_list_EN.csv │ ├── size_list_CN.csv │ ├── size_list_EN.csv │ └── title.md ├── config.py ├── images │ ├── test0.jpg │ ├── test1.jpg │ ├── test2.jpg │ ├── test3.jpg │ └── test4.jpg ├── locales.py ├── processor.py ├── ui.py └── utils.py ├── deploy_api.py ├── docker-compose.yml ├── docs ├── api_CN.md ├── api_EN.md ├── face++_CN.md └── face++_EN.md ├── hivision ├── __init__.py ├── creator │ ├── __init__.py │ ├── choose_handler.py │ ├── context.py │ ├── face_detector.py │ ├── human_matting.py │ ├── layout_calculator.py │ ├── move_image.py │ ├── photo_adjuster.py │ ├── retinaface │ │ ├── __init__.py │ │ ├── box_utils.py │ │ ├── inference.py │ │ ├── prior_box.py │ │ └── weights │ │ │ └── .gitkeep │ ├── rotation_adjust.py │ ├── tensor2numpy.py │ ├── utils.py │ └── weights │ │ └── .gitkeep ├── error.py ├── plugin │ ├── beauty │ │ ├── __init__.py │ │ ├── base_adjust.py │ │ ├── beauty_tools.py │ │ ├── grind_skin.py │ │ ├── handler.py │ │ ├── lut │ │ │ └── lut_origin.png │ │ ├── thin_face.py │ │ └── whitening.py │ ├── font │ │ ├── .gitkeep │ │ └── 青鸟华光简琥珀.ttf │ ├── template │ │ ├── assets │ │ │ ├── template_1.png │ │ │ ├── template_2.png │ │ │ └── template_config.json │ │ └── template_calculator.py │ └── watermark.py └── utils.py ├── inference.py ├── requirements-app.txt ├── requirements-dev.txt ├── requirements.txt ├── scripts ├── build_pypi.py └── download_model.py └── test ├── create_id_photo.py └── temp └── .gitkeep /.devcontainer/devcontainer.json: -------------------------------------------------------------------------------- 1 | // For format details, see https://aka.ms/devcontainer.json. For config options, see the 2 | // README at: https://github.com/devcontainers/templates/tree/main/src/universal 3 | { 4 | "name": "Default Linux Universal", 5 | // Or use a Dockerfile or Docker Compose file. More info: https://containers.dev/guide/dockerfile 6 | "image": "mcr.microsoft.com/devcontainers/universal:2-linux", 7 | 8 | // Features to add to the dev container. More info: https://containers.dev/features. 9 | // "features": {}, 10 | 11 | // Use 'forwardPorts' to make a list of ports inside the container available locally. 12 | // "forwardPorts": [], 13 | 14 | // Use 'postCreateCommand' to run commands after the container is created. 15 | "onCreateCommand": "sh .devcontainer/start.sh", 16 | 17 | // Configure tool-specific properties. 18 | "customizations": { 19 | "vscode": { 20 | "extensions": [ 21 | "ms-python.python", 22 | "eamodio.gitlens", 23 | "mhutchie.git-graph" 24 | ] 25 | } 26 | } 27 | 28 | // Uncomment to connect as root instead. More info: https://aka.ms/dev-containers-non-root. 29 | // "remoteUser": "root" 30 | } 31 | -------------------------------------------------------------------------------- /.devcontainer/start.sh: -------------------------------------------------------------------------------- 1 | sudo apt-get update && sudo apt-get install ffmpeg libsm6 libxext6 -y 2 | 3 | conda create -n HivisionIDPhotos python=3.10 -y 4 | conda init 5 | echo 'conda activate HivisionIDPhotos' >> ~/.bashrc 6 | 7 | ENV_PATH="/opt/conda/envs/HivisionIDPhotos/bin" 8 | $ENV_PATH/pip install -r requirements.txt -r requirements-app.txt -r requirements-dev.txt 9 | 10 | $ENV_PATH/python scripts/download_model.py --models all 11 | -------------------------------------------------------------------------------- /.github/workflows/build-image.yml: -------------------------------------------------------------------------------- 1 | name: build image and push 2 | 3 | on: 4 | push: 5 | tags: 6 | - '*' 7 | 8 | jobs: 9 | docker: 10 | runs-on: ubuntu-latest 11 | environment: release 12 | steps: 13 | - name: Checkout 14 | uses: actions/checkout@v4 15 | 16 | - name: Set up Python 17 | uses: actions/setup-python@v4 18 | with: 19 | python-version: '3.10' 20 | 21 | - name: Install dependencies 22 | run: pip install requests tqdm 23 | 24 | - name: Download models 25 | run: python scripts/download_model.py --models all 26 | 27 | - name: Set up QEMU 28 | uses: docker/setup-qemu-action@v3 29 | 30 | - name: Set up Docker Buildx 31 | uses: docker/setup-buildx-action@v3 32 | 33 | - name: Login to Docker Hub 34 | uses: docker/login-action@v3 35 | with: 36 | username: ${{ vars.DOCKERHUB_USERNAME }} 37 | password: ${{ secrets.DOCKERHUB_TOKEN }} 38 | 39 | - name: Build and push 40 | uses: docker/build-push-action@v6 41 | with: 42 | context: . 43 | platforms: linux/amd64,linux/arm64 44 | push: true 45 | tags: | 46 | ${{ vars.IMAGE_NAME }}:latest 47 | ${{ vars.IMAGE_NAME }}:${{ github.ref_name }} -------------------------------------------------------------------------------- /.gitignore: -------------------------------------------------------------------------------- 1 | *.pyc 2 | **/__pycache__/ 3 | .idea 4 | .vscode/* 5 | .history 6 | .DS_Store 7 | .env 8 | demo/kb_output/*.jpg 9 | demo/kb_output/*.png 10 | scripts/sync_swanhub.py 11 | scripts/sync_huggingface.py 12 | scripts/sync_modelscope.py 13 | scripts/sync_all.py 14 | **/flagged/ 15 | # build outputs 16 | dist 17 | build 18 | # checkpoint 19 | *.pth 20 | *.pt 21 | *.onnx 22 | *.mnn 23 | test/temp/* 24 | !test/temp/.gitkeep 25 | 26 | .python-version 27 | 28 | # Ignore .png and .jpg files in the root directory 29 | /*.png 30 | /*.jpg 31 | -------------------------------------------------------------------------------- /.vscode/extensions.json: -------------------------------------------------------------------------------- 1 | { 2 | "recommendations": [ 3 | "ms-python.black-formatter", 4 | "donjayamanne.python-extension-pack", 5 | "njpwerner.autodocstring", 6 | 7 | "editorconfig.editorconfig", 8 | 9 | "gruntfuggly.todo-tree", 10 | 11 | "eamodio.gitlens", 12 | 13 | "PKief.material-icon-theme", 14 | "davidanson.vscode-markdownlint", 15 | "usernamehw.errorlens", 16 | "tamasfe.even-better-toml", 17 | 18 | "littlefoxteam.vscode-python-test-adapter" 19 | ] 20 | } 21 | -------------------------------------------------------------------------------- /.vscode/settings.json: -------------------------------------------------------------------------------- 1 | { 2 | "workbench.iconTheme": "material-icon-theme", 3 | "material-icon-theme.files.associations": { 4 | ".env.mock": "Tune", 5 | "requirements-dev.txt": "python-misc", 6 | "requirements-media.txt": "python-misc" 7 | }, 8 | /** 后端代码格式化部分，python格式化 */ 9 | "[python]": { 10 | "editor.defaultFormatter": "ms-python.black-formatter", 11 | "editor.formatOnSave": true 12 | }, 13 | /** TODO tree 配置 */ 14 | "todo-tree.general.tags": [ 15 | "TODO", // 待办 16 | "FIXME", // 待修复 17 | "COMPAT", // 兼容性问题 18 | "WARNING" // 警告 19 | ], 20 | "todo-tree.highlights.customHighlight": { 21 | "TODO": { 22 | "icon": "check", 23 | "type": "tag", 24 | "foreground": "#ffff00", 25 | "iconColour": "#ffff" 26 | }, 27 | "WARNING": { 28 | "icon": "alert", 29 | "type": "tag", 30 | "foreground": "#ff0000", 31 | "iconColour": "#ff0000" 32 | }, 33 | "FIXME": { 34 | "icon": "flame", 35 | "type": "tag", 36 | "foreground": "#ff0000", 37 | "iconColour": "#ff0000" 38 | }, 39 | "COMPAT": { 40 | "icon": "flame", 41 | "type": "tag", 42 | "foreground": "#00ff00", 43 | "iconColour": "#ffff" 44 | } 45 | }, 46 | 47 | /** python代码注释 */ 48 | "autoDocstring.docstringFormat": "numpy", 49 | 50 | /** markdown格式检查 */ 51 | "markdownlint.config": { 52 | // 允许使用html标签 53 | "MD033": false, 54 | // 允许首行不是level1标题 55 | "MD041": false 56 | }, 57 | 58 | /** 不显示文件夹 */ 59 | "files.exclude": { 60 | "**/.git": true, 61 | "**/.svn": true, 62 | "**/.hg": true, 63 | "**/CVS": true, 64 | "**/.DS_Store": true, 65 | "**/Thumbs.db": true, 66 | "**/__pycache__": true, 67 | ".idea": true 68 | }, 69 | "python.testing.pytestEnabled": true, 70 | "ros.distro": "humble" 71 | } 72 | -------------------------------------------------------------------------------- /Dockerfile: -------------------------------------------------------------------------------- 1 | FROM python:3.10-slim 2 | 3 | # Install system dependencies 4 | RUN apt-get update && apt-get install -y --no-install-recommends \ 5 | ffmpeg \ 6 | libgl1-mesa-glx \ 7 | libglib2.0-0 \ 8 | && rm -rf /var/lib/apt/lists/* 9 | 10 | WORKDIR /app 11 | 12 | COPY requirements.txt requirements-app.txt ./ 13 | 14 | RUN pip install --no-cache-dir -r requirements.txt -r requirements-app.txt 15 | 16 | COPY . . 17 | 18 | EXPOSE 7860 19 | EXPOSE 8080 20 | 21 | CMD ["python3", "-u", "app.py", "--host", "0.0.0.0", "--port", "7860"] -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | Apache License 2 | Version 2.0, January 2004 3 | http://www.apache.org/licenses/ 4 | 5 | TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 6 | 7 | 1. Definitions. 8 | 9 | "License" shall mean the terms and conditions for use, reproduction, 10 | and distribution as defined by Sections 1 through 9 of this document. 11 | 12 | "Licensor" shall mean the copyright owner or entity authorized by 13 | the copyright owner that is granting the License. 14 | 15 | "Legal Entity" shall mean the union of the acting entity and all 16 | other entities that control, are controlled by, or are under common 17 | control with that entity. For the purposes of this definition, 18 | "control" means (i) the power, direct or indirect, to cause the 19 | direction or management of such entity, whether by contract or 20 | otherwise, or (ii) ownership of fifty percent (50%) or more of the 21 | outstanding shares, or (iii) beneficial ownership of such entity. 22 | 23 | "You" (or "Your") shall mean an individual or Legal Entity 24 | exercising permissions granted by this License. 25 | 26 | "Source" form shall mean the preferred form for making modifications, 27 | including but not limited to software source code, documentation 28 | source, and configuration files. 29 | 30 | "Object" form shall mean any form resulting from mechanical 31 | transformation or translation of a Source form, including but 32 | not limited to compiled object code, generated documentation, 33 | and conversions to other media types. 34 | 35 | "Work" shall mean the work of authorship, whether in Source or 36 | Object form, made available under the License, as indicated by a 37 | copyright notice that is included in or attached to the work 38 | (an example is provided in the Appendix below). 39 | 40 | "Derivative Works" shall mean any work, whether in Source or Object 41 | form, that is based on (or derived from) the Work and for which the 42 | editorial revisions, annotations, elaborations, or other modifications 43 | represent, as a whole, an original work of authorship. For the purposes 44 | of this License, Derivative Works shall not include works that remain 45 | separable from, or merely link (or bind by name) to the interfaces of, 46 | the Work and Derivative Works thereof. 47 | 48 | "Contribution" shall mean any work of authorship, including 49 | the original version of the Work and any modifications or additions 50 | to that Work or Derivative Works thereof, that is intentionally 51 | submitted to Licensor for inclusion in the Work by the copyright owner 52 | or by an individual or Legal Entity authorized to submit on behalf of 53 | the copyright owner. For the purposes of this definition, "submitted" 54 | means any form of electronic, verbal, or written communication sent 55 | to the Licensor or its representatives, including but not limited to 56 | communication on electronic mailing lists, source code control systems, 57 | and issue tracking systems that are managed by, or on behalf of, the 58 | Licensor for the purpose of discussing and improving the Work, but 59 | excluding communication that is conspicuously marked or otherwise 60 | designated in writing by the copyright owner as "Not a Contribution." 61 | 62 | "Contributor" shall mean Licensor and any individual or Legal Entity 63 | on behalf of whom a Contribution has been received by Licensor and 64 | subsequently incorporated within the Work. 65 | 66 | 2. Grant of Copyright License. Subject to the terms and conditions of 67 | this License, each Contributor hereby grants to You a perpetual, 68 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 69 | copyright license to reproduce, prepare Derivative Works of, 70 | publicly display, publicly perform, sublicense, and distribute the 71 | Work and such Derivative Works in Source or Object form. 72 | 73 | 3. Grant of Patent License. Subject to the terms and conditions of 74 | this License, each Contributor hereby grants to You a perpetual, 75 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 76 | (except as stated in this section) patent license to make, have made, 77 | use, offer to sell, sell, import, and otherwise transfer the Work, 78 | where such license applies only to those patent claims licensable 79 | by such Contributor that are necessarily infringed by their 80 | Contribution(s) alone or by combination of their Contribution(s) 81 | with the Work to which such Contribution(s) was submitted. If You 82 | institute patent litigation against any entity (including a 83 | cross-claim or counterclaim in a lawsuit) alleging that the Work 84 | or a Contribution incorporated within the Work constitutes direct 85 | or contributory patent infringement, then any patent licenses 86 | granted to You under this License for that Work shall terminate 87 | as of the date such litigation is filed. 88 | 89 | 4. Redistribution. You may reproduce and distribute copies of the 90 | Work or Derivative Works thereof in any medium, with or without 91 | modifications, and in Source or Object form, provided that You 92 | meet the following conditions: 93 | 94 | (a) You must give any other recipients of the Work or 95 | Derivative Works a copy of this License; and 96 | 97 | (b) You must cause any modified files to carry prominent notices 98 | stating that You changed the files; and 99 | 100 | (c) You must retain, in the Source form of any Derivative Works 101 | that You distribute, all copyright, patent, trademark, and 102 | attribution notices from the Source form of the Work, 103 | excluding those notices that do not pertain to any part of 104 | the Derivative Works; and 105 | 106 | (d) If the Work includes a "NOTICE" text file as part of its 107 | distribution, then any Derivative Works that You distribute must 108 | include a readable copy of the attribution notices contained 109 | within such NOTICE file, excluding those notices that do not 110 | pertain to any part of the Derivative Works, in at least one 111 | of the following places: within a NOTICE text file distributed 112 | as part of the Derivative Works; within the Source form or 113 | documentation, if provided along with the Derivative Works; or, 114 | within a display generated by the Derivative Works, if and 115 | wherever such third-party notices normally appear. The contents 116 | of the NOTICE file are for informational purposes only and 117 | do not modify the License. You may add Your own attribution 118 | notices within Derivative Works that You distribute, alongside 119 | or as an addendum to the NOTICE text from the Work, provided 120 | that such additional attribution notices cannot be construed 121 | as modifying the License. 122 | 123 | You may add Your own copyright statement to Your modifications and 124 | may provide additional or different license terms and conditions 125 | for use, reproduction, or distribution of Your modifications, or 126 | for any such Derivative Works as a whole, provided Your use, 127 | reproduction, and distribution of the Work otherwise complies with 128 | the conditions stated in this License. 129 | 130 | 5. Submission of Contributions. Unless You explicitly state otherwise, 131 | any Contribution intentionally submitted for inclusion in the Work 132 | by You to the Licensor shall be under the terms and conditions of 133 | this License, without any additional terms or conditions. 134 | Notwithstanding the above, nothing herein shall supersede or modify 135 | the terms of any separate license agreement you may have executed 136 | with Licensor regarding such Contributions. 137 | 138 | 6. Trademarks. This License does not grant permission to use the trade 139 | names, trademarks, service marks, or product names of the Licensor, 140 | except as required for reasonable and customary use in describing the 141 | origin of the Work and reproducing the content of the NOTICE file. 142 | 143 | 7. Disclaimer of Warranty. Unless required by applicable law or 144 | agreed to in writing, Licensor provides the Work (and each 145 | Contributor provides its Contributions) on an "AS IS" BASIS, 146 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 147 | implied, including, without limitation, any warranties or conditions 148 | of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A 149 | PARTICULAR PURPOSE. You are solely responsible for determining the 150 | appropriateness of using or redistributing the Work and assume any 151 | risks associated with Your exercise of permissions under this License. 152 | 153 | 8. Limitation of Liability. In no event and under no legal theory, 154 | whether in tort (including negligence), contract, or otherwise, 155 | unless required by applicable law (such as deliberate and grossly 156 | negligent acts) or agreed to in writing, shall any Contributor be 157 | liable to You for damages, including any direct, indirect, special, 158 | incidental, or consequential damages of any character arising as a 159 | result of this License or out of the use or inability to use the 160 | Work (including but not limited to damages for loss of goodwill, 161 | work stoppage, computer failure or malfunction, or any and all 162 | other commercial damages or losses), even if such Contributor 163 | has been advised of the possibility of such damages. 164 | 165 | 9. Accepting Warranty or Additional Liability. While redistributing 166 | the Work or Derivative Works thereof, You may choose to offer, 167 | and charge a fee for, acceptance of support, warranty, indemnity, 168 | or other liability obligations and/or rights consistent with this 169 | License. However, in accepting such obligations, You may act only 170 | on Your own behalf and on Your sole responsibility, not on behalf 171 | of any other Contributor, and only if You agree to indemnify, 172 | defend, and hold each Contributor harmless for any liability 173 | incurred by, or claims asserted against, such Contributor by reason 174 | of your accepting any such warranty or additional liability. 175 | 176 | END OF TERMS AND CONDITIONS 177 | 178 | APPENDIX: How to apply the Apache License to your work. 179 | 180 | To apply the Apache License to your work, attach the following 181 | boilerplate notice, with the fields enclosed by brackets "[]" 182 | replaced with your own identifying information. (Don't include 183 | the brackets!) The text should be enclosed in the appropriate 184 | comment syntax for the file format. We also recommend that a 185 | file or class name and description of purpose be included on the 186 | same "printed page" as the copyright notice for easier 187 | identification within third-party archives. 188 | 189 | Copyright [yyyy] [name of copyright owner] 190 | 191 | Licensed under the Apache License, Version 2.0 (the "License"); 192 | you may not use this file except in compliance with the License. 193 | You may obtain a copy of the License at 194 | 195 | http://www.apache.org/licenses/LICENSE-2.0 196 | 197 | Unless required by applicable law or agreed to in writing, software 198 | distributed under the License is distributed on an "AS IS" BASIS, 199 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 200 | See the License for the specific language governing permissions and 201 | limitations under the License. -------------------------------------------------------------------------------- /app.py: -------------------------------------------------------------------------------- 1 | import argparse 2 | import os 3 | from demo.processor import IDPhotoProcessor 4 | from demo.ui import create_ui 5 | from hivision.creator.choose_handler import HUMAN_MATTING_MODELS 6 | 7 | root_dir = os.path.dirname(os.path.abspath(__file__)) 8 | 9 | # 获取存在的人像分割模型列表 10 | # 通过检查 hivision/creator/weights 目录下的 .onnx 和 .mnn 文件 11 | # 只保留文件名（不包括扩展名） 12 | HUMAN_MATTING_MODELS_EXIST = [ 13 | os.path.splitext(file)[0] 14 | for file in os.listdir(os.path.join(root_dir, "hivision/creator/weights")) 15 | if file.endswith(".onnx") or file.endswith(".mnn") 16 | ] 17 | # 在HUMAN_MATTING_MODELS中的模型才会被加载到Gradio中显示 18 | HUMAN_MATTING_MODELS_CHOICE = [ 19 | model for model in HUMAN_MATTING_MODELS if model in HUMAN_MATTING_MODELS_EXIST 20 | ] 21 | 22 | if len(HUMAN_MATTING_MODELS_CHOICE) == 0: 23 | raise ValueError( 24 | "未找到任何存在的人像分割模型，请检查 hivision/creator/weights 目录下的文件" 25 | + "\n" 26 | + "No existing portrait segmentation model was found, please check the files in the hivision/creator/weights directory." 27 | ) 28 | 29 | FACE_DETECT_MODELS = ["face++ (联网Online API)", "mtcnn"] 30 | FACE_DETECT_MODELS_EXPAND = ( 31 | ["retinaface-resnet50"] 32 | if os.path.exists( 33 | os.path.join( 34 | root_dir, "hivision/creator/retinaface/weights/retinaface-resnet50.onnx" 35 | ) 36 | ) 37 | else [] 38 | ) 39 | FACE_DETECT_MODELS_CHOICE = FACE_DETECT_MODELS + FACE_DETECT_MODELS_EXPAND 40 | 41 | LANGUAGE = ["zh", "en", "ko", "ja"] 42 | 43 | if __name__ == "__main__": 44 | argparser = argparse.ArgumentParser() 45 | argparser.add_argument( 46 | "--port", type=int, default=7860, help="The port number of the server" 47 | ) 48 | argparser.add_argument( 49 | "--host", type=str, default="127.0.0.1", help="The host of the server" 50 | ) 51 | argparser.add_argument( 52 | "--root_path", 53 | type=str, 54 | default=None, 55 | help="The root path of the server, default is None (='/'), e.g. '/myapp'", 56 | ) 57 | args = argparser.parse_args() 58 | 59 | processor = IDPhotoProcessor() 60 | 61 | demo = create_ui( 62 | processor, 63 | root_dir, 64 | HUMAN_MATTING_MODELS_CHOICE, 65 | FACE_DETECT_MODELS_CHOICE, 66 | LANGUAGE, 67 | ) 68 | 69 | # 如果RUN_MODE是Beast，打印已开启野兽模式 70 | if os.getenv("RUN_MODE") == "beast": 71 | print("[Beast mode activated.] 已开启野兽模式。") 72 | 73 | demo.launch( 74 | server_name=args.host, 75 | server_port=args.port, 76 | favicon_path=os.path.join(root_dir, "assets/hivision_logo.png"), 77 | root_path=args.root_path, 78 | show_api=False, 79 | ) 80 | -------------------------------------------------------------------------------- /app.spec: -------------------------------------------------------------------------------- 1 | # -*- mode: python ; coding: utf-8 -*- 2 | from PyInstaller.utils.hooks import collect_data_files 3 | 4 | datas = [('hivisionai', 'hivisionai'), ('hivision_modnet.onnx', '.'), ('size_list_CN.csv', '.')] 5 | datas += collect_data_files('gradio_client') 6 | datas += collect_data_files('gradio') 7 | 8 | 9 | a = Analysis( 10 | ['app/web.py'], 11 | pathex=[], 12 | binaries=[], 13 | datas=datas, 14 | hiddenimports=[], 15 | hookspath=[], 16 | hooksconfig={}, 17 | runtime_hooks=[], 18 | excludes=[], 19 | noarchive=False, 20 | optimize=0, 21 | ) 22 | pyz = PYZ(a.pure) 23 | 24 | exe = EXE( 25 | pyz, 26 | a.scripts, 27 | a.binaries, 28 | a.datas, 29 | [], 30 | name='HivisionIDPhotos', 31 | debug=False, 32 | bootloader_ignore_signals=False, 33 | strip=False, 34 | upx=True, 35 | upx_exclude=[], 36 | runtime_tmpdir=None, 37 | console=True, 38 | disable_windowed_traceback=False, 39 | argv_emulation=False, 40 | target_arch=None, 41 | codesign_identity=None, 42 | entitlements_file=None, 43 | icon=['assets\hivisionai.ico'], 44 | ) 45 | -------------------------------------------------------------------------------- /assets/comfyui.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/assets/comfyui.png -------------------------------------------------------------------------------- /assets/community-uniapp-wechat-miniprogram.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/assets/community-uniapp-wechat-miniprogram.png -------------------------------------------------------------------------------- /assets/community-web.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/assets/community-web.png -------------------------------------------------------------------------------- /assets/community-wechat-miniprogram.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/assets/community-wechat-miniprogram.png -------------------------------------------------------------------------------- /assets/demo.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/assets/demo.png -------------------------------------------------------------------------------- /assets/demoImage.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/assets/demoImage.jpg -------------------------------------------------------------------------------- /assets/face++.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/assets/face++.png -------------------------------------------------------------------------------- /assets/gradio-image.jpeg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/assets/gradio-image.jpeg -------------------------------------------------------------------------------- /assets/harry.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/assets/harry.png -------------------------------------------------------------------------------- /assets/hivision_logo.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/assets/hivision_logo.png -------------------------------------------------------------------------------- /assets/social_template.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/assets/social_template.png -------------------------------------------------------------------------------- /demo/assets/american-style.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/demo/assets/american-style.png -------------------------------------------------------------------------------- /demo/assets/color_list_CN.csv: -------------------------------------------------------------------------------- 1 | Name,Hex 2 | 蓝色,628bce 3 | 白色,ffffff 4 | 红色,d74532 5 | 黑色,000000 6 | 深蓝色,4b6190 7 | 浅灰色,f2f0f0 -------------------------------------------------------------------------------- /demo/assets/color_list_EN.csv: -------------------------------------------------------------------------------- 1 | Name,Hex 2 | Blue,628bce 3 | White,ffffff 4 | Red,d74532 5 | Black,000000 6 | Dark Blue,4b6190 7 | Light Gray,f2f0f0 -------------------------------------------------------------------------------- /demo/assets/size_list_CN.csv: -------------------------------------------------------------------------------- 1 | Name,Height,Width 2 | 一寸,413,295 3 | 二寸,626,413 4 | 小一寸,378,260 5 | 小二寸,531,413 6 | 大一寸,567,390 7 | 大二寸,626,413 8 | 五寸,1499,1050 9 | 教师资格证,413,295 10 | 国家公务员考试,413,295 11 | 初级会计考试,413,295 12 | 英语四六级考试,192,144 13 | 计算机等级考试,567,390 14 | 研究生考试,709,531 15 | 社保卡,441,358 16 | 电子驾驶证,378,260 17 | 美国签证,600,600 18 | 日本签证,413,295 19 | 韩国签证,531,413 -------------------------------------------------------------------------------- /demo/assets/size_list_EN.csv: -------------------------------------------------------------------------------- 1 | Name,Height,Width 2 | One inch,413,295 3 | Two inches,626,413 4 | Small one inch,378,260 5 | Small two inches,531,413 6 | Large one inch,567,390 7 | Large two inches,626,413 8 | Five inches,1499,1050 9 | Teacher qualification certificate,413,295 10 | National civil service exa,413,295 11 | Primary accounting exam,413,295 12 | English CET-4 and CET-6 exams,192,144 13 | Computer level exam,567,390 14 | Graduate entrance exam,709,531 15 | Social security card,441,358 16 | Electronic driver's license,378,260 17 | American visa,600,600 18 | Japanese visa,413,295 19 | Korean visa,531,413 -------------------------------------------------------------------------------- /demo/assets/title.md: -------------------------------------------------------------------------------- 1 |
2 |
3 | HivisionIDPhotos 4 | HivisionIDPhotos v1.3.1 5 |
6 |
7 | Github   8 | GitHub stars   9 | SwanLab   10 | Static Badge 11 |
12 |
13 | -------------------------------------------------------------------------------- /demo/config.py: -------------------------------------------------------------------------------- 1 | import os 2 | from demo.utils import csv_to_size_list, csv_to_color_list 3 | 4 | 5 | def load_configuration(root_dir): 6 | size_list_dict_CN = csv_to_size_list( 7 | os.path.join(root_dir, "assets/size_list_CN.csv") 8 | ) 9 | size_list_dict_EN = csv_to_size_list( 10 | os.path.join(root_dir, "assets/size_list_EN.csv") 11 | ) 12 | color_list_dict_CN = csv_to_color_list( 13 | os.path.join(root_dir, "assets/color_list_CN.csv") 14 | ) 15 | color_list_dict_EN = csv_to_color_list( 16 | os.path.join(root_dir, "assets/color_list_EN.csv") 17 | ) 18 | 19 | return size_list_dict_CN, size_list_dict_EN, color_list_dict_CN, color_list_dict_EN 20 | -------------------------------------------------------------------------------- /demo/images/test0.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/demo/images/test0.jpg -------------------------------------------------------------------------------- /demo/images/test1.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/demo/images/test1.jpg -------------------------------------------------------------------------------- /demo/images/test2.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/demo/images/test2.jpg -------------------------------------------------------------------------------- /demo/images/test3.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/demo/images/test3.jpg -------------------------------------------------------------------------------- /demo/images/test4.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/demo/images/test4.jpg -------------------------------------------------------------------------------- /demo/utils.py: -------------------------------------------------------------------------------- 1 | import csv 2 | 3 | 4 | def csv_to_size_list(csv_file: str) -> dict: 5 | # 初始化一个空字典 6 | size_list_dict = {} 7 | 8 | # 打开 CSV 文件并读取数据 9 | with open(csv_file, mode="r", encoding="utf-8") as file: 10 | reader = csv.reader(file) 11 | # 跳过表头 12 | next(reader) 13 | # 读取数据并填充字典 14 | for row in reader: 15 | size_name, h, w = row 16 | size_name_add_size = "{}\t\t({}, {})".format(size_name, h, w) 17 | size_list_dict[size_name_add_size] = (int(h), int(w)) 18 | 19 | return size_list_dict 20 | 21 | 22 | def csv_to_color_list(csv_file: str) -> dict: 23 | # 初始化一个空字典 24 | color_list_dict = {} 25 | 26 | # 打开 CSV 文件并读取数据 27 | with open(csv_file, mode="r", encoding="utf-8") as file: 28 | reader = csv.reader(file) 29 | # 跳过表头 30 | next(reader) 31 | # 读取数据并填充字典 32 | for row in reader: 33 | color_name, hex_code = row 34 | color_list_dict[color_name] = hex_code 35 | 36 | return color_list_dict 37 | 38 | 39 | def range_check(value, min_value=0, max_value=255): 40 | value = int(value) 41 | return max(min_value, min(value, max_value)) 42 | -------------------------------------------------------------------------------- /deploy_api.py: -------------------------------------------------------------------------------- 1 | from fastapi import FastAPI, UploadFile, Form, File 2 | from hivision import IDCreator 3 | from hivision.error import FaceError 4 | from hivision.creator.layout_calculator import ( 5 | generate_layout_array, 6 | generate_layout_image, 7 | ) 8 | from hivision.creator.choose_handler import choose_handler 9 | from hivision.utils import ( 10 | add_background, 11 | resize_image_to_kb, 12 | bytes_2_base64, 13 | base64_2_numpy, 14 | hex_to_rgb, 15 | add_watermark, 16 | save_image_dpi_to_bytes, 17 | ) 18 | import numpy as np 19 | import cv2 20 | from starlette.middleware.cors import CORSMiddleware 21 | from starlette.formparsers import MultiPartParser 22 | 23 | # 设置Starlette表单字段大小限制 24 | MultiPartParser.max_part_size = 10 * 1024 * 1024 # 10MB 25 | # 设置Starlette文件上传大小限制 26 | MultiPartParser.max_file_size = 20 * 1024 * 1024 # 20MB 27 | 28 | app = FastAPI() 29 | creator = IDCreator() 30 | 31 | # 添加 CORS 中间件 解决跨域问题 32 | app.add_middleware( 33 | CORSMiddleware, 34 | allow_origins=["*"], # 允许的请求来源 35 | allow_credentials=True, # 允许携带 Cookie 36 | allow_methods=[ 37 | "*" 38 | ], # 允许的请求方法，例如：GET, POST 等，也可以指定 ["GET", "POST"] 39 | allow_headers=["*"], # 允许的请求头，也可以指定具体的头部 40 | ) 41 | 42 | 43 | # 证件照智能制作接口 44 | @app.post("/idphoto") 45 | async def idphoto_inference( 46 | input_image: UploadFile = File(None), 47 | input_image_base64: str = Form(None), 48 | height: int = Form(413), 49 | width: int = Form(295), 50 | human_matting_model: str = Form("modnet_photographic_portrait_matting"), 51 | face_detect_model: str = Form("mtcnn"), 52 | hd: bool = Form(True), 53 | dpi: int = Form(300), 54 | face_align: bool = Form(False), 55 | head_measure_ratio: float = Form(0.2), 56 | head_height_ratio: float = Form(0.45), 57 | top_distance_max: float = Form(0.12), 58 | top_distance_min: float = Form(0.10), 59 | brightness_strength: float = Form(0), 60 | contrast_strength: float = Form(0), 61 | sharpen_strength: float = Form(0), 62 | saturation_strength: float = Form(0), 63 | ): 64 | # 如果传入了base64，则直接使用base64解码 65 | if input_image_base64: 66 | img = base64_2_numpy(input_image_base64) 67 | # 否则使用上传的图片 68 | else: 69 | image_bytes = await input_image.read() 70 | nparr = np.frombuffer(image_bytes, np.uint8) 71 | img = cv2.imdecode(nparr, cv2.IMREAD_COLOR) 72 | 73 | # ------------------- 选择抠图与人脸检测模型 ------------------- 74 | choose_handler(creator, human_matting_model, face_detect_model) 75 | 76 | # 将字符串转为元组 77 | size = (int(height), int(width)) 78 | try: 79 | result = creator( 80 | img, 81 | size=size, 82 | head_measure_ratio=head_measure_ratio, 83 | head_height_ratio=head_height_ratio, 84 | head_top_range=(top_distance_max, top_distance_min), 85 | face_alignment=face_align, 86 | brightness_strength=brightness_strength, 87 | contrast_strength=contrast_strength, 88 | sharpen_strength=sharpen_strength, 89 | saturation_strength=saturation_strength, 90 | ) 91 | except FaceError: 92 | result_message = {"status": False} 93 | # 如果检测到人脸数量等于1, 则返回标准证和高清照结果（png 4通道图像） 94 | else: 95 | result_image_standard_bytes = save_image_dpi_to_bytes(cv2.cvtColor(result.standard, cv2.COLOR_RGBA2BGRA), None, dpi) 96 | 97 | result_message = { 98 | "status": True, 99 | "image_base64_standard": bytes_2_base64(result_image_standard_bytes), 100 | } 101 | 102 | # 如果hd为True, 则增加高清照结果（png 4通道图像） 103 | if hd: 104 | result_image_hd_bytes = save_image_dpi_to_bytes(cv2.cvtColor(result.hd, cv2.COLOR_RGBA2BGRA), None, dpi) 105 | result_message["image_base64_hd"] = bytes_2_base64(result_image_hd_bytes) 106 | 107 | return result_message 108 | 109 | 110 | # 人像抠图接口 111 | @app.post("/human_matting") 112 | async def human_matting_inference( 113 | input_image: UploadFile = File(None), 114 | input_image_base64: str = Form(None), 115 | human_matting_model: str = Form("hivision_modnet"), 116 | dpi: int = Form(300), 117 | ): 118 | if input_image_base64: 119 | img = base64_2_numpy(input_image_base64) 120 | else: 121 | image_bytes = await input_image.read() 122 | nparr = np.frombuffer(image_bytes, np.uint8) 123 | img = cv2.imdecode(nparr, cv2.IMREAD_COLOR) 124 | 125 | # ------------------- 选择抠图与人脸检测模型 ------------------- 126 | choose_handler(creator, human_matting_model, None) 127 | 128 | try: 129 | result = creator( 130 | img, 131 | change_bg_only=True, 132 | ) 133 | except FaceError: 134 | result_message = {"status": False} 135 | 136 | else: 137 | result_image_standard_bytes = save_image_dpi_to_bytes(cv2.cvtColor(result.standard, cv2.COLOR_RGBA2BGRA), None, dpi) 138 | result_message = { 139 | "status": True, 140 | "image_base64": bytes_2_base64(result_image_standard_bytes), 141 | } 142 | return result_message 143 | 144 | 145 | # 透明图像添加纯色背景接口 146 | @app.post("/add_background") 147 | async def photo_add_background( 148 | input_image: UploadFile = File(None), 149 | input_image_base64: str = Form(None), 150 | color: str = Form("000000"), 151 | kb: int = Form(None), 152 | dpi: int = Form(300), 153 | render: int = Form(0), 154 | ): 155 | render_choice = ["pure_color", "updown_gradient", "center_gradient"] 156 | 157 | if input_image_base64: 158 | img = base64_2_numpy(input_image_base64) 159 | else: 160 | image_bytes = await input_image.read() 161 | nparr = np.frombuffer(image_bytes, np.uint8) 162 | img = cv2.imdecode(nparr, cv2.IMREAD_UNCHANGED) 163 | 164 | color = hex_to_rgb(color) 165 | color = (color[2], color[1], color[0]) 166 | 167 | result_image = add_background( 168 | img, 169 | bgr=color, 170 | mode=render_choice[render], 171 | ).astype(np.uint8) 172 | 173 | result_image = cv2.cvtColor(result_image, cv2.COLOR_RGB2BGR) 174 | if kb: 175 | result_image_bytes = resize_image_to_kb(result_image, None, int(kb), dpi=dpi) 176 | else: 177 | result_image_bytes = save_image_dpi_to_bytes(result_image, None, dpi=dpi) 178 | 179 | result_messgae = { 180 | "status": True, 181 | "image_base64": bytes_2_base64(result_image_bytes), 182 | } 183 | 184 | return result_messgae 185 | 186 | 187 | # 六寸排版照生成接口 188 | @app.post("/generate_layout_photos") 189 | async def generate_layout_photos( 190 | input_image: UploadFile = File(None), 191 | input_image_base64: str = Form(None), 192 | height: int = Form(413), 193 | width: int = Form(295), 194 | kb: int = Form(None), 195 | dpi: int = Form(300), 196 | ): 197 | # try: 198 | if input_image_base64: 199 | img = base64_2_numpy(input_image_base64) 200 | else: 201 | image_bytes = await input_image.read() 202 | nparr = np.frombuffer(image_bytes, np.uint8) 203 | img = cv2.imdecode(nparr, cv2.IMREAD_COLOR) 204 | 205 | size = (int(height), int(width)) 206 | 207 | typography_arr, typography_rotate = generate_layout_array( 208 | input_height=size[0], input_width=size[1] 209 | ) 210 | 211 | result_layout_image = generate_layout_image( 212 | img, typography_arr, typography_rotate, height=size[0], width=size[1] 213 | ).astype(np.uint8) 214 | 215 | result_layout_image = cv2.cvtColor(result_layout_image, cv2.COLOR_RGB2BGR) 216 | if kb: 217 | result_layout_image_bytes = resize_image_to_kb( 218 | result_layout_image, None, int(kb), dpi=dpi 219 | ) 220 | else: 221 | result_layout_image_bytes = save_image_dpi_to_bytes(result_layout_image, None, dpi=dpi) 222 | 223 | result_layout_image_base64 = bytes_2_base64(result_layout_image_bytes) 224 | 225 | result_messgae = { 226 | "status": True, 227 | "image_base64": result_layout_image_base64, 228 | } 229 | 230 | return result_messgae 231 | 232 | 233 | # 透明图像添加水印接口 234 | @app.post("/watermark") 235 | async def watermark( 236 | input_image: UploadFile = File(None), 237 | input_image_base64: str = Form(None), 238 | text: str = Form("Hello"), 239 | size: int = 20, 240 | opacity: float = 0.5, 241 | angle: int = 30, 242 | color: str = "#000000", 243 | space: int = 25, 244 | kb: int = Form(None), 245 | dpi: int = Form(300), 246 | ): 247 | if input_image_base64: 248 | img = base64_2_numpy(input_image_base64) 249 | else: 250 | image_bytes = await input_image.read() 251 | nparr = np.frombuffer(image_bytes, np.uint8) 252 | img = cv2.imdecode(nparr, cv2.IMREAD_COLOR) 253 | 254 | try: 255 | result_image = add_watermark(img, text, size, opacity, angle, color, space) 256 | 257 | result_image = cv2.cvtColor(result_image, cv2.COLOR_RGB2BGR) 258 | if kb: 259 | result_image_bytes = resize_image_to_kb(result_image, None, int(kb), dpi=dpi) 260 | else: 261 | result_image_bytes = save_image_dpi_to_bytes(result_image, None, dpi=dpi) 262 | result_image_base64 = bytes_2_base64(result_image_bytes) 263 | 264 | result_messgae = { 265 | "status": True, 266 | "image_base64": result_image_base64, 267 | } 268 | except Exception as e: 269 | result_messgae = { 270 | "status": False, 271 | "error": str(e), 272 | } 273 | 274 | return result_messgae 275 | 276 | 277 | # 设置照片KB值接口(RGB图) 278 | @app.post("/set_kb") 279 | async def set_kb( 280 | input_image: UploadFile = File(None), 281 | input_image_base64: str = Form(None), 282 | dpi: int = Form(300), 283 | kb: int = Form(50), 284 | ): 285 | if input_image_base64: 286 | img = base64_2_numpy(input_image_base64) 287 | else: 288 | image_bytes = await input_image.read() 289 | nparr = np.frombuffer(image_bytes, np.uint8) 290 | img = cv2.imdecode(nparr, cv2.IMREAD_COLOR) 291 | 292 | try: 293 | result_image = cv2.cvtColor(img, cv2.COLOR_RGB2BGR) 294 | result_image_bytes = resize_image_to_kb(result_image, None, int(kb), dpi=dpi) 295 | result_image_base64 = bytes_2_base64(result_image_bytes) 296 | 297 | result_messgae = { 298 | "status": True, 299 | "image_base64": result_image_base64, 300 | } 301 | except Exception as e: 302 | result_messgae = { 303 | "status": False, 304 | "error": e, 305 | } 306 | 307 | return result_messgae 308 | 309 | 310 | # 证件照智能裁剪接口 311 | @app.post("/idphoto_crop") 312 | async def idphoto_crop_inference( 313 | input_image: UploadFile = File(None), 314 | input_image_base64: str = Form(None), 315 | height: int = Form(413), 316 | width: int = Form(295), 317 | face_detect_model: str = Form("mtcnn"), 318 | hd: bool = Form(True), 319 | dpi: int = Form(300), 320 | head_measure_ratio: float = Form(0.2), 321 | head_height_ratio: float = Form(0.45), 322 | top_distance_max: float = Form(0.12), 323 | top_distance_min: float = Form(0.10), 324 | ): 325 | if input_image_base64: 326 | img = base64_2_numpy(input_image_base64) 327 | else: 328 | image_bytes = await input_image.read() 329 | nparr = np.frombuffer(image_bytes, np.uint8) 330 | img = cv2.imdecode(nparr, cv2.IMREAD_UNCHANGED) # 读取图像(4通道) 331 | 332 | # ------------------- 选择抠图与人脸检测模型 ------------------- 333 | choose_handler(creator, face_detect_option=face_detect_model) 334 | 335 | # 将字符串转为元组 336 | size = (int(height), int(width)) 337 | try: 338 | result = creator( 339 | img, 340 | size=size, 341 | head_measure_ratio=head_measure_ratio, 342 | head_height_ratio=head_height_ratio, 343 | head_top_range=(top_distance_max, top_distance_min), 344 | crop_only=True, 345 | ) 346 | except FaceError: 347 | result_message = {"status": False} 348 | # 如果检测到人脸数量等于1, 则返回标准证和高清照结果（png 4通道图像） 349 | else: 350 | result_image_standard_bytes = save_image_dpi_to_bytes(cv2.cvtColor(result.standard, cv2.COLOR_RGBA2BGRA), None, dpi) 351 | 352 | result_message = { 353 | "status": True, 354 | "image_base64_standard": bytes_2_base64(result_image_standard_bytes), 355 | } 356 | 357 | # 如果hd为True, 则增加高清照结果（png 4通道图像） 358 | if hd: 359 | result_image_hd_bytes = save_image_dpi_to_bytes(cv2.cvtColor(result.hd, cv2.COLOR_RGBA2BGRA), None, dpi) 360 | result_message["image_base64_hd"] = bytes_2_base64(result_image_hd_bytes) 361 | 362 | return result_message 363 | 364 | 365 | if __name__ == "__main__": 366 | import uvicorn 367 | 368 | # 在8080端口运行推理服务 369 | uvicorn.run(app, host="0.0.0.0", port=8080) 370 | -------------------------------------------------------------------------------- /docker-compose.yml: -------------------------------------------------------------------------------- 1 | version: '3.8' 2 | 3 | services: 4 | hivision_idphotos: 5 | build: 6 | context: . 7 | dockerfile: Dockerfile 8 | image: linzeyi/hivision_idphotos 9 | command: python3 -u app.py --host 0.0.0.0 --port 7860 10 | ports: 11 | - '7860:7860' 12 | 13 | hivision_idphotos_api: 14 | build: 15 | context: . 16 | dockerfile: Dockerfile 17 | image: linzeyi/hivision_idphotos 18 | command: python3 deploy_api.py 19 | ports: 20 | - '8080:8080' 21 | -------------------------------------------------------------------------------- /docs/api_CN.md: -------------------------------------------------------------------------------- 1 | # API Docs 2 | 3 | [English](api_EN.md) / 中文 4 | 5 | 6 | ## 目录 7 | 8 | - [开始之前：开启后端服务](#开始之前开启后端服务) 9 | - [接口功能说明](#接口功能说明) 10 | - [1.生成证件照(底透明)](#1生成证件照底透明) 11 | - [2.添加背景色](#2添加背景色) 12 | - [3.生成六寸排版照](#3生成六寸排版照) 13 | - [4.人像抠图](#4人像抠图) 14 | - [5.图像加水印](#5图像加水印) 15 | - [6.设置图像KB大小](#6设置图像KB大小) 16 | - [7.证件照裁切](#7证件照裁切) 17 | - [cURL 请求示例](#curl-请求示例) 18 | - [Python 请求示例](#python-请求示例) 19 | 20 | ## 开始之前：开启后端服务 21 | 22 | 在请求 API 之前，请先运行后端服务 23 | 24 | ```bash 25 | python deploy_api.py 26 | ``` 27 | 28 |
29 | 30 | ## 接口功能说明 31 | 32 | ### 1.生成证件照(底透明) 33 | 34 | 接口名：`idphoto` 35 | 36 | `生成证件照`接口的逻辑是发送一张 RGB 图像，输出一张标准证件照和一张高清证件照： 37 | 38 | - **高清证件照**：根据`size`的宽高比例制作的证件照，文件名为`output_image_dir`增加`_hd`后缀 39 | - **标准证件照**：尺寸等于`size`，由高清证件照缩放而来，文件名为`output_image_dir` 40 | 41 | 需要注意的是，生成的两张照片都是透明的（RGBA 四通道图像），要生成完整的证件照，还需要下面的`添加背景色`接口。 42 | 43 | > 问：为什么这么设计？ 44 | > 答：因为在实际产品中，经常用户会频繁切换底色预览效果，直接给透明底图像，由前端 js 代码合成颜色是更好体验的做法。 45 | 46 | **请求参数：** 47 | 48 | | 参数名 | 类型 | 必填 | 说明 | 49 | | :--- | :--- | :--- | :--- | 50 | | input_image | file | 和`input_image_base64`二选一 | 传入的图像文件，图像文件为需为RGB三通道图像。 | 51 | | input_image_base64 | str | 和`input_image`二选一 | 传入的图像文件的base64编码，图像文件为需为RGB三通道图像。 | 52 | | height | int | 否 | 标准证件照高度，默认为`413` | 53 | | width | int | 否 | 标准证件照宽度，默认为`295` | 54 | | human_matting_model | str | 否 | 人像分割模型，默认为`modnet_photographic_portrait_matting`。可选值为`modnet_photographic_portrait_matting`、`hivision_modnet`、`rmbg-1.4`、`birefnet-v1-lite` | 55 | | face_detect_model | str | 否 | 人脸检测模型，默认为`mtcnn`。可选值为`mtcnn`、`face_plusplus`、`retinaface-resnet50` | 56 | | hd | bool | 否 | 是否生成高清证件照，默认为`true` | 57 | | dpi | int | 否 | 图像分辨率，默认为`300` | 58 | | face_alignment | bool | 否 | 是否进行人脸对齐，默认为`true` | 59 | | head_measure_ratio | float | 否 | 面部面积与照片面积的比例，默认为`0.2` | 60 | | head_height_ratio | float | 否 | 面部中心与照片顶部的高度比例，默认为`0.45` | 61 | | top_distance_max | float | 否 | 头部与照片顶部距离的比例最大值，默认为`0.12` | 62 | | top_distance_min | float | 否 | 头部与照片顶部距离的比例最小值，默认为`0.1` | 63 | | brightness_strength | float | 否 | 亮度调整强度，默认为`0` | 64 | | contrast_strength | float | 否 | 对比度调整强度，默认为`0` | 65 | | sharpen_strength | float | 否 | 锐化调整强度，默认为`0` | 66 | | saturation_strength | float | 否 | 饱和度调整强度，默认为`0` | 67 | 68 | **返回参数：** 69 | 70 | | 参数名 | 类型 | 说明 | 71 | | :--- | :--- | :--- | 72 | | status | int | 状态码，`true`表示成功 | 73 | | image_base64_standard | str | 标准证件照的base64编码 | 74 | | image_base64_hd | str | 高清证件照的base64编码。如`hd`参数为`false`，则不返回该参数 | 75 | 76 |
77 | 78 | ### 2.添加背景色 79 | 80 | 接口名：`add_background` 81 | 82 | `添加背景色`接口的逻辑是接收一张 RGBA 图像（透明图），根据`color`添加背景色，合成一张 JPG 图像。 83 | 84 | **请求参数：** 85 | 86 | | 参数名 | 类型 | 必填 | 说明 | 87 | | :--- | :--- | :--- | :--- | 88 | | input_image | file | 和`input_image_base64`二选一 | 传入的图像文件，图像文件为需为RGBA四通道图像。 | 89 | | input_image_base64 | str | 和`input_image`二选一 | 传入的图像文件的base64编码，图像文件为需为RGBA四通道图像。 | 90 | | color | str | 否 | 背景色HEX值，默认为`000000` | 91 | | kb | int | 否 | 输出照片的 KB 值，默认为`None`，即不对图像进行KB调整。| 92 | | render | int | 否 | 渲染模式，默认为`0`。可选值为`0`、`1`、`2`，分别对应`纯色`、`上下渐变`、`中心渐变`。 | 93 | | dpi | int | 否 | 图像分辨率，默认为`300` | 94 | 95 | **返回参数：** 96 | 97 | | 参数名 | 类型 | 说明 | 98 | | :--- | :--- | :--- | 99 | | status | int | 状态码，`true`表示成功 | 100 | | image_base64 | str | 添加背景色之后的图像的base64编码 | 101 | 102 |
103 | 104 | ### 3.生成六寸排版照 105 | 106 | 接口名：`generate_layout_photos` 107 | 108 | `生成六寸排版照`接口的逻辑是接收一张 RGB 图像（一般为添加背景色之后的证件照），根据`size`进行照片排布，然后生成一张六寸排版照。 109 | 110 | **请求参数：** 111 | 112 | | 参数名 | 类型 | 必填 | 说明 | 113 | | :--- | :--- | :--- | :--- | 114 | | input_image | file | 和`input_image_base64`二选一 | 传入的图像文件，图像文件为需为RGB三通道图像。 | 115 | | input_image_base64 | str | 和`input_image`二选一 | 传入的图像文件的base64编码，图像文件为需为RGB三通道图像。 | 116 | | height | int | 否 | 输入图像的高度，默认为`413` | 117 | | width | int | 否 | 输入图像的宽度，默认为`295` | 118 | | kb | int | 否 | 输出照片的 KB 值，默认为`None`，即不对图像进行KB调整。| 119 | | dpi | int | 否 | 图像分辨率，默认为`300` | 120 | 121 | **返回参数：** 122 | 123 | | 参数名 | 类型 | 说明 | 124 | | :--- | :--- | :--- | 125 | | status | int | 状态码，`true`表示成功 | 126 | | image_base64 | str | 六寸排版照的base64编码 | 127 | 128 |
129 | 130 | ### 4.人像抠图 131 | 132 | 接口名：`human_matting` 133 | 134 | `人像抠图`接口的逻辑是接收一张 RGB 图像，输出一张标准抠图人像照和高清抠图人像照（无任何背景填充）。 135 | 136 | **请求参数：** 137 | 138 | | 参数名 | 类型 | 必填 | 说明 | 139 | | :--- | :--- | :--- | :--- | 140 | | input_image | file | 是 | 传入的图像文件，图像文件为需为RGB三通道图像。 | 141 | | human_matting_model | str | 否 | 人像分割模型，默认为`modnet_photographic_portrait_matting`。可选值为`modnet_photographic_portrait_matting`、`hivision_modnet`、`rmbg-1.4`、`birefnet-v1-lite` | 142 | | dpi | int | 否 | 图像分辨率，默认为`300` | 143 | 144 | **返回参数：** 145 | 146 | | 参数名 | 类型 | 说明 | 147 | | :--- | :--- | :--- | 148 | | status | int | 状态码，`true`表示成功 | 149 | | image_base64 | str | 抠图人像照的base64编码 | 150 | 151 |
152 | 153 | ### 5.图像加水印 154 | 155 | 接口名：`watermark` 156 | 157 | `图像加水印`接口的功能是接收一个水印文本，然后在原图上添加指定的水印。用户可以指定水印的位置、透明度和大小等属性，以便将水印无缝地融合到原图中。 158 | 159 | **请求参数：** 160 | 161 | | 参数名 | 类型 | 必填 | 说明 | 162 | | :--- | :--- | :--- | :--- | 163 | | input_image | file | 和`input_image_base64`二选一 | 传入的图像文件，图像文件为需为RGB三通道图像。 | 164 | | input_image_base64 | str | 和`input_image`二选一 | 传入的图像文件的base64编码，图像文件为需为RGB三通道图像。 | 165 | | text | str | 否 | 水印文本，默认为`Hello` | 166 | | size | int | 否 | 水印字体大小，默认为`20` | 167 | | opacity | float | 否 | 水印透明度，默认为`0.5` | 168 | | angle | int | 否 | 水印旋转角度，默认为`30` | 169 | | color | str | 否 | 水印颜色，默认为`#000000` | 170 | | space | int | 否 | 水印间距，默认为`25` | 171 | | dpi | int | 否 | 图像分辨率，默认为`300` | 172 | 173 | **返回参数：** 174 | 175 | | 参数名 | 类型 | 说明 | 176 | | :--- | :--- | :--- | 177 | | status | int | 状态码，`true`表示成功 | 178 | | image_base64 | str | 添加水印之后的图像的base64编码 | 179 | 180 |
181 | 182 | ### 6.设置图像KB大小 183 | 184 | 接口名：`set_kb` 185 | 186 | `设置图像KB大小`接口的功能是接收一张图像和目标文件大小（以KB为单位），如果设置的KB值小于原文件，则调整压缩率；如果设置的KB值大于源文件，则通过给文件头添加信息的方式调大KB值，目标是让图像的最终大小与设置的KB值一致。 187 | 188 | **请求参数：** 189 | 190 | | 参数名 | 类型 | 必填 | 说明 | 191 | | :--- | :--- | :--- | :--- | 192 | | input_image | file | 和`input_image_base64`二选一 | 传入的图像文件，图像文件为需为RGB三通道图像。 | 193 | | input_image_base64 | str | 和`input_image`二选一 | 传入的图像文件的base64编码，图像文件为需为RGB三通道图像。 | 194 | | kb | int | 否 | 输出照片的 KB 值，默认为`None`，即不对图像进行KB调整。| 195 | | dpi | int | 否 | 图像分辨率，默认为`300` | 196 | 197 | **返回参数：** 198 | 199 | | 参数名 | 类型 | 说明 | 200 | | :--- | :--- | :--- | 201 | | status | int | 状态码，`true`表示成功 | 202 | | image_base64 | str | 设置KB大小之后的图像的base64编码 | 203 | 204 |
205 | 206 | ### 7.证件照裁切 207 | 208 | 接口名：`idphoto_crop` 209 | 210 | `证件照裁切`接口的功能是接收一张 RBGA 图像（透明图），输出一张标准证件照和一张高清证件照。 211 | 212 | **请求参数：** 213 | 214 | | 参数名 | 类型 | 必填 | 说明 | 215 | | :--- | :--- | :--- | :--- | 216 | | input_image | file | 和`input_image_base64`二选一 | 传入的图像文件，图像文件为需为RGBA四通道图像。 | 217 | | input_image_base64 | str | 和`input_image`二选一 | 传入的图像文件的base64编码，图像文件为需为RGBA四通道图像。 | 218 | | height | int | 否 | 标准证件照高度，默认为`413` | 219 | | width | int | 否 | 标准证件照宽度，默认为`295` | 220 | | face_detect_model | str | 否 | 人脸检测模型，默认为`mtcnn`。可选值为`mtcnn`、`face_plusplus`、`retinaface-resnet50` | 221 | | hd | bool | 否 | 是否生成高清证件照，默认为`true` | 222 | | dpi | int | 否 | 图像分辨率，默认为`300` | 223 | | head_measure_ratio | float | 否 | 面部面积与照片面积的比例，默认为`0.2` | 224 | | head_height_ratio | float | 否 | 面部中心与照片顶部的高度比例，默认为`0.45` | 225 | | top_distance_max | float | 否 | 头部与照片顶部距离的比例最大值，默认为`0.12` | 226 | | top_distance_min | float | 否 | 头部与照片顶部距离的比例最小值，默认为`0.1` | 227 | 228 | **返回参数：** 229 | 230 | | 参数名 | 类型 | 说明 | 231 | | :--- | :--- | :--- | 232 | | status | int | 状态码，`true`表示成功 | 233 | | image_base64 | str | 证件照裁切之后的图像的base64编码 | 234 | | image_base64_hd | str | 高清证件照裁切之后的图像的base64编码，如`hd`参数为`false`，则不返回该参数 | 235 | 236 |
237 | 238 | ## cURL 请求示例 239 | 240 | cURL 是一个命令行工具，用于使用各种网络协议传输数据。以下是使用 cURL 调用这些 API 的示例。 241 | 242 | ### 1. 生成证件照(底透明) 243 | 244 | ```bash 245 | curl -X POST "http://127.0.0.1:8080/idphoto" \ 246 | -F "input_image=@demo/images/test0.jpg" \ 247 | -F "height=413" \ 248 | -F "width=295" \ 249 | -F "human_matting_model=modnet_photographic_portrait_matting" \ 250 | -F "face_detect_model=mtcnn" \ 251 | -F "hd=true" \ 252 | -F "dpi=300" \ 253 | -F "face_alignment=true" \ 254 | -F 'head_height_ratio=0.45' \ 255 | -F 'head_measure_ratio=0.2' \ 256 | -F 'top_distance_min=0.1' \ 257 | -F 'top_distance_max=0.12' \ 258 | -F 'sharpen_strength=0' \ 259 | -F 'saturation_strength=0' \ 260 | -F 'brightness_strength=10' \ 261 | -F 'contrast_strength=0' 262 | ``` 263 | 264 | ### 2. 添加背景色 265 | 266 | ```bash 267 | curl -X POST "http://127.0.0.1:8080/add_background" \ 268 | -F "input_image=@test.png" \ 269 | -F "color=638cce" \ 270 | -F "kb=200" \ 271 | -F "render=0" \ 272 | -F "dpi=300" 273 | ``` 274 | 275 | ### 3. 生成六寸排版照 276 | 277 | ```bash 278 | curl -X POST "http://127.0.0.1:8080/generate_layout_photos" \ 279 | -F "input_image=@test.jpg" \ 280 | -F "height=413" \ 281 | -F "width=295" \ 282 | -F "kb=200" \ 283 | -F "dpi=300" 284 | ``` 285 | 286 | ### 4. 人像抠图 287 | 288 | ```bash 289 | curl -X POST "http://127.0.0.1:8080/human_matting" \ 290 | -F "input_image=@demo/images/test0.jpg" \ 291 | -F "human_matting_model=modnet_photographic_portrait_matting" \ 292 | -F "dpi=300" 293 | ``` 294 | 295 | ### 5. 图片加水印 296 | ```bash 297 | curl -X 'POST' \ 298 | 'http://127.0.0.1:8080/watermark?size=20&opacity=0.5&angle=30&color=%23000000&space=25' \ 299 | -H 'accept: application/json' \ 300 | -H 'Content-Type: multipart/form-data' \ 301 | -F 'input_image=@demo/images/test0.jpg;type=image/jpeg' \ 302 | -F 'text=Hello' \ 303 | -F 'dpi=300' 304 | ``` 305 | 306 | ### 6. 设置图像KB大小 307 | ```bash 308 | curl -X 'POST' \ 309 | 'http://127.0.0.1:8080/set_kb' \ 310 | -H 'accept: application/json' \ 311 | -H 'Content-Type: multipart/form-data' \ 312 | -F 'input_image=@demo/images/test0.jpg;type=image/jpeg' \ 313 | -F 'kb=50' \ 314 | -F 'dpi=300' 315 | ``` 316 | 317 | ### 7. 证件照裁切 318 | ```bash 319 | curl -X 'POST' 'http://127.0.0.1:8080/idphoto_crop' \ 320 | -H 'accept: application/json' \ 321 | -H 'Content-Type: multipart/form-data' \ 322 | -F 'input_image=@idphoto_matting.png;type=image/png' \ 323 | -F 'height=413' \ 324 | -F 'width=295' \ 325 | -F 'face_detect_model=mtcnn' \ 326 | -F 'hd=true' \ 327 | -F 'dpi=300' \ 328 | -F 'head_height_ratio=0.45' \ 329 | -F 'head_measure_ratio=0.2' \ 330 | -F 'top_distance_min=0.1' \ 331 | -F 'top_distance_max=0.12' 332 | ``` 333 | 334 |
335 | 336 | ## Python 请求示例 337 | 338 | #### 1.生成证件照(底透明) 339 | ```python 340 | import requests 341 | 342 | url = "http://127.0.0.1:8080/idphoto" 343 | input_image_path = "demo/images/test0.jpg" 344 | 345 | files = {"input_image": open(input_image_path, "rb")} 346 | data = { 347 | "height": 413, 348 | "width": 295, 349 | "human_matting_model": "modnet_photographic_portrait_matting", 350 | "face_detect_model": "mtcnn", 351 | "hd": True, 352 | "dpi": 300, 353 | "face_alignment": True, 354 | "head_measure_ratio": 0.2, 355 | "head_height_ratio": 0.45, 356 | "top_distance_max": 0.12, 357 | "top_distance_min": 0.1, 358 | "brightness_strength": 0, 359 | "contrast_strength": 0, 360 | "sharpen_strength": 0, 361 | "saturation_strength": 0, 362 | } 363 | 364 | response = requests.post(url, params=params, files=files, data=data).json() 365 | 366 | # response为一个json格式字典，包含status、image_base64_standard和image_base64_hd三项 367 | print(response) 368 | ``` 369 | 370 | #### 2.添加背景色 371 | 372 | ```python 373 | import requests 374 | 375 | url = "http://127.0.0.1:8080/add_background" 376 | input_image_path = "test.png" 377 | 378 | files = {"input_image": open(input_image_path, "rb")} 379 | data = { 380 | "color": '638cce', 381 | "kb": None, 382 | "render": 0, 383 | "dpi": 300, 384 | } 385 | 386 | response = requests.post(url, files=files, data=data).json() 387 | 388 | # response为一个json格式字典，包含status和image_base64 389 | print(response) 390 | ``` 391 | 392 | #### 3.生成六寸排版照 393 | 394 | ```python 395 | import requests 396 | 397 | url = "http://127.0.0.1:8080/generate_layout_photos" 398 | input_image_path = "test.jpg" 399 | 400 | files = {"input_image": open(input_image_path, "rb")} 401 | data = { 402 | "height": 413, 403 | "width": 295, 404 | "kb": 200, 405 | "dpi": 300, 406 | } 407 | 408 | response = requests.post(url, files=files, data=data).json() 409 | 410 | # response为一个json格式字典，包含status和image_base64 411 | print(response) 412 | ``` 413 | 414 | #### 4.人像抠图 415 | 416 | ```python 417 | import requests 418 | 419 | url = "http://127.0.0.1:8080/human_matting" 420 | input_image_path = "test.jpg" 421 | 422 | files = {"input_image": open(input_image_path, "rb")} 423 | data = { 424 | "human_matting_model": "modnet_photographic_portrait_matting", 425 | "dpi": 300, 426 | } 427 | 428 | response = requests.post(url, files=files, data=data).json() 429 | 430 | # response为一个json格式字典，包含status和image_base64 431 | print(response) 432 | ``` 433 | 434 | #### 5.图片加水印 435 | 436 | ```python 437 | import requests 438 | 439 | # 设置请求的 URL 和参数 440 | url = "http://127.0.0.1:8080/watermark" 441 | params = { 442 | "size": 20, 443 | "opacity": 0.5, 444 | "angle": 30, 445 | "color": "#000000", 446 | "space": 25, 447 | } 448 | 449 | # 设置文件和其他表单数据 450 | input_image_path = "demo/images/test0.jpg" 451 | files = {"input_image": open(input_image_path, "rb")} 452 | data = {"text": "Hello", "dpi": 300} 453 | 454 | # 发送 POST 请求 455 | response = requests.post(url, params=params, files=files, data=data) 456 | 457 | # 检查响应 458 | if response.ok: 459 | # 输出响应内容 460 | print(response.json()) 461 | else: 462 | # 输出错误信息 463 | print(f"Request failed with status code {response.status_code}: {response.text}") 464 | ``` 465 | 466 | ### 6. 设置图像KB大小 467 | 468 | ```python 469 | import requests 470 | 471 | # 设置请求的 URL 472 | url = "http://127.0.0.1:8080/set_kb" 473 | 474 | # 设置文件和其他表单数据 475 | input_image_path = "demo/images/test0.jpg" 476 | files = {"input_image": open(input_image_path, "rb")} 477 | data = {"kb": 50, "dpi": 300} 478 | 479 | # 发送 POST 请求 480 | response = requests.post(url, files=files, data=data) 481 | 482 | # 检查响应 483 | if response.ok: 484 | # 输出响应内容 485 | print(response.json()) 486 | else: 487 | # 输出错误信息 488 | print(f"Request failed with status code {response.status_code}: {response.text}") 489 | ``` 490 | 491 | ### 7. 证件照裁切 492 | 493 | ```python 494 | import requests 495 | 496 | # 设置请求的 URL 497 | url = "http://127.0.0.1:8080/idphoto_crop" 498 | 499 | # 设置文件和其他表单数据 500 | input_image_path = "idphoto_matting.png" 501 | files = {"input_image": ("idphoto_matting.png", open(input_image_path, "rb"), "image/png")} 502 | data = { 503 | "height": 413, 504 | "width": 295, 505 | "face_detect_model": "mtcnn", 506 | "hd": "true", 507 | "dpi": 300, 508 | "head_measure_ratio": 0.2, 509 | "head_height_ratio": 0.45, 510 | "top_distance_max": 0.12, 511 | "top_distance_min": 0.1, 512 | } 513 | 514 | # 发送 POST 请求 515 | response = requests.post(url, files=files, data=data) 516 | 517 | # 检查响应 518 | if response.ok: 519 | # 输出响应内容 520 | print(response.json()) 521 | else: 522 | # 输出错误信息 523 | print(f"Request failed with status code {response.status_code}: {response.text}") 524 | ``` -------------------------------------------------------------------------------- /docs/face++_CN.md: -------------------------------------------------------------------------------- 1 | # Face++ 人脸检测配置文档 2 | 3 | [Face++官方文档](https://console.faceplusplus.com.cn/documents/4888373) 4 | 5 | ## 1. 注册Face++账号 6 | 要使用 Face++ 的人脸检测 API，您首先需要在 Face++ 的官方网站上注册一个账号。注册后，您将能够访问 API 控制台和相关服务。 7 | 8 | ### 步骤： 9 | 1. 访问 [Face++ 官网](https://www.faceplusplus.com.cn/)。 10 | 2. 点击“注册”按钮，填写相关信息以创建您的账号。 11 | 12 | ## 2. 获取API KEY和API SECRET 13 | 注册并登录后，您需要获取用于身份验证的 API Key 和 API Secret。这些信息是调用 API 时必需的。 14 | 15 | ### 步骤： 16 | 1. 登录到您的 Face++ 账号。 17 | 2. 进入 控制台 -> 应用管理 -> API Key。 18 | 3. 在控制台中，您将看到您的 API Key 和 API Secret。 19 | 20 | ## 3. 设置环境变量 21 | 为了在代码中安全地使用 API Key 和 API Secret，建议将它们设置为环境变量。这样可以避免在代码中硬编码敏感信息。 22 | 23 | ### 在不同操作系统中设置环境变量的步骤： 24 | - **Windows**: 25 | 1. 打开命令提示符。 26 | 2. 输入以下命令并按回车： 27 | ```cmd 28 | set FACE_PLUS_API_KEY="您的API_KEY" 29 | set FACE_PLUS_API_SECRET="您的API_SECRET" 30 | ``` 31 | 32 | - **Linux / macOS**: 33 | 1. 打开终端。 34 | 2. 输入以下命令并按回车： 35 | ```bash 36 | export FACE_PLUS_API_KEY="你的API_KEY" 37 | export FACE_PLUS_API_SECRET="你的API_SECRET" 38 | ``` 39 | 40 | > **注意**: 您可能需要在启动应用程序之前运行上述命令，或者将这些命令添加到您的 shell 配置文件（例如 `.bashrc` 或 `.bash_profile`）中，以便每次启动终端时自动加载。 41 | 42 | ## 4. 启动Gradio服务 43 | 44 | 运行gradio服务，在「人脸检测模型」中选择「face++」即可。 45 | 46 | ```bash 47 | python app.py 48 | ``` 49 | 50 | ![alt text](../assets/face++.png) 51 | 52 | ## 错误码的解释 53 | 54 | https://console.faceplusplus.com.cn/documents/4888373 -------------------------------------------------------------------------------- /docs/face++_EN.md: -------------------------------------------------------------------------------- 1 | Here's the translated document in English: 2 | 3 | # Face++ Face Detection Configuration Document 4 | 5 | [Face++ Official Documentation](https://console.faceplusplus.com.cn/documents/4888373) 6 | 7 | ## 1. Register a Face++ Account 8 | To use the Face++ Face Detection API, you first need to register an account on the Face++ official website. After registration, you will be able to access the API console and related services. 9 | 10 | ### Steps: 11 | 1. Visit the [Face++ Official Website](https://www.faceplusplus.com.cn/). 12 | 2. Click the "Register" button and fill in the relevant information to create your account. 13 | 14 | ## 2. Obtain API KEY and API SECRET 15 | After registering and logging in, you need to obtain the API Key and API Secret for authentication. This information is necessary for calling the API. 16 | 17 | ### Steps: 18 | 1. Log in to your Face++ account. 19 | 2. Go to Console -> Application Management -> API Key. 20 | 3. In the console, you will see your API Key and API Secret. 21 | 22 | ## 3. Set Environment Variables 23 | To securely use the API Key and API Secret in your code, it is recommended to set them as environment variables. This avoids hardcoding sensitive information in your code. 24 | 25 | ### Steps to Set Environment Variables on Different Operating Systems: 26 | - **Windows**: 27 | 1. Open the Command Prompt. 28 | 2. Enter the following commands and press Enter: 29 | ```cmd 30 | set FACE_PLUS_API_KEY="Your_API_KEY" 31 | set FACE_PLUS_API_SECRET="Your_API_SECRET" 32 | ``` 33 | 34 | - **Linux / macOS**: 35 | 1. Open the terminal. 36 | 2. Enter the following commands and press Enter: 37 | ```bash 38 | export FACE_PLUS_API_KEY="Your_API_KEY" 39 | export FACE_PLUS_API_SECRET="Your_API_SECRET" 40 | ``` 41 | 42 | > **Note**: You may need to run the above commands before starting your application, or add these commands to your shell configuration file (e.g., `.bashrc` or `.bash_profile`) so that they are automatically loaded each time you start the terminal. 43 | 44 | ## 4. Start Gradio Service 45 | Run the Gradio service, and select "face++" in the "Face Detection Model". 46 | 47 | ```bash 48 | python app.py 49 | ``` 50 | 51 | ![alt text](../assets/face++.png) 52 | 53 | ## Explanation of error codes 54 | 55 | https://console.faceplusplus.com.cn/documents/4888373 -------------------------------------------------------------------------------- /hivision/__init__.py: -------------------------------------------------------------------------------- 1 | from .creator import IDCreator, Params as IDParams, Result as IDResult 2 | 3 | 4 | __all__ = ["IDCreator", "IDParams", "IDResult", "utils", "error"] 5 | -------------------------------------------------------------------------------- /hivision/creator/__init__.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding: utf-8 -*- 3 | r""" 4 | @DATE: 2024/9/5 16:45 5 | @File: __init__.py 6 | @IDE: pycharm 7 | @Description: 8 | 创建证件照 9 | """ 10 | import numpy as np 11 | from typing import Tuple 12 | import hivision.creator.utils as U 13 | from .context import Context, ContextHandler, Params, Result 14 | from .human_matting import extract_human 15 | from .face_detector import detect_face_mtcnn 16 | from hivision.plugin.beauty.handler import beauty_face 17 | from .photo_adjuster import adjust_photo 18 | import cv2 19 | import time 20 | 21 | 22 | class IDCreator: 23 | """ 24 | 证件照创建类，包含完整的证件照流程 25 | """ 26 | 27 | def __init__(self): 28 | # 回调时机 29 | self.before_all: ContextHandler = None 30 | """ 31 | 在所有处理之前，此时图像已经被 resize 到最大边长为 2000 32 | """ 33 | self.after_matting: ContextHandler = None 34 | """ 35 | 在抠图之后，ctx.matting_image 被赋值 36 | """ 37 | self.after_detect: ContextHandler = None 38 | """ 39 | 在人脸检测之后，ctx.face 被赋值，如果为仅换底，则不会执行此回调 40 | """ 41 | self.after_all: ContextHandler = None 42 | """ 43 | 在所有处理之后，此时 ctx.result 被赋值 44 | """ 45 | # 处理者 46 | self.matting_handler: ContextHandler = extract_human 47 | self.detection_handler: ContextHandler = detect_face_mtcnn 48 | self.beauty_handler: ContextHandler = beauty_face 49 | # 上下文 50 | self.ctx = None 51 | 52 | def __call__( 53 | self, 54 | image: np.ndarray, 55 | size: Tuple[int, int] = (413, 295), 56 | change_bg_only: bool = False, 57 | crop_only: bool = False, 58 | head_measure_ratio: float = 0.2, 59 | head_height_ratio: float = 0.45, 60 | head_top_range: float = (0.12, 0.1), 61 | face: Tuple[int, int, int, int] = None, 62 | whitening_strength: int = 0, 63 | brightness_strength: int = 0, 64 | contrast_strength: int = 0, 65 | sharpen_strength: int = 0, 66 | saturation_strength: int = 0, 67 | face_alignment: bool = False, 68 | ) -> Result: 69 | """ 70 | 证件照处理函数 71 | :param image: 输入图像 72 | :param change_bg_only: 是否只需要抠图 73 | :param crop_only: 是否只需要裁剪 74 | :param size: 输出的图像大小（h,w) 75 | :param head_measure_ratio: 人脸面积与全图面积的期望比值 76 | :param head_height_ratio: 人脸中心处在全图高度的比例期望值 77 | :param head_top_range: 头距离顶部的比例（max,min) 78 | :param face: 人脸坐标 79 | :param whitening_strength: 美白强度 80 | :param brightness_strength: 亮度强度 81 | :param contrast_strength: 对比度强度 82 | :param sharpen_strength: 锐化强度 83 | :param align_face: 是否需要人脸矫正 84 | 85 | :return: 返回处理后的证件照和一系列参数 86 | """ 87 | # 0.初始化上下文 88 | params = Params( 89 | size=size, 90 | change_bg_only=change_bg_only, 91 | head_measure_ratio=head_measure_ratio, 92 | head_height_ratio=head_height_ratio, 93 | head_top_range=head_top_range, 94 | crop_only=crop_only, 95 | face=face, 96 | whitening_strength=whitening_strength, 97 | brightness_strength=brightness_strength, 98 | contrast_strength=contrast_strength, 99 | sharpen_strength=sharpen_strength, 100 | saturation_strength=saturation_strength, 101 | face_alignment=face_alignment, 102 | ) 103 | 104 | 105 | # 总的开始时间 106 | total_start_time = time.time() 107 | 108 | self.ctx = Context(params) 109 | ctx = self.ctx 110 | ctx.processing_image = image 111 | ctx.processing_image = U.resize_image_esp( 112 | ctx.processing_image, 2000 113 | ) # 将输入图片 resize 到最大边长为 2000 114 | ctx.origin_image = ctx.processing_image.copy() 115 | self.before_all and self.before_all(ctx) 116 | 117 | # 1. ------------------人像抠图------------------ 118 | # 如果仅裁剪，则不进行抠图 119 | if not ctx.params.crop_only: 120 | # 调用抠图工作流 121 | print("[1] Start Human Matting...") 122 | start_matting_time = time.time() 123 | self.matting_handler(ctx) 124 | end_matting_time = time.time() 125 | print(f"[1] Human Matting Time: {end_matting_time - start_matting_time:.3f}s") 126 | self.after_matting and self.after_matting(ctx) 127 | # 如果进行抠图 128 | else: 129 | ctx.matting_image = ctx.processing_image 130 | 131 | 132 | # 2. ------------------美颜------------------ 133 | print("[2] Start Beauty...") 134 | start_beauty_time = time.time() 135 | self.beauty_handler(ctx) 136 | end_beauty_time = time.time() 137 | print(f"[2] Beauty Time: {end_beauty_time - start_beauty_time:.3f}s") 138 | 139 | # 如果仅换底，则直接返回抠图结果 140 | if ctx.params.change_bg_only: 141 | ctx.result = Result( 142 | standard=ctx.matting_image, 143 | hd=ctx.matting_image, 144 | matting=ctx.matting_image, 145 | clothing_params=None, 146 | typography_params=None, 147 | face=None, 148 | ) 149 | self.after_all and self.after_all(ctx) 150 | return ctx.result 151 | 152 | # 3. ------------------人脸检测------------------ 153 | print("[3] Start Face Detection...") 154 | start_detection_time = time.time() 155 | self.detection_handler(ctx) 156 | end_detection_time = time.time() 157 | print(f"[3] Face Detection Time: {end_detection_time - start_detection_time:.3f}s") 158 | self.after_detect and self.after_detect(ctx) 159 | 160 | # 3.1 ------------------人脸对齐------------------ 161 | if ctx.params.face_alignment and abs(ctx.face["roll_angle"]) > 2: 162 | print("[3.1] Start Face Alignment...") 163 | start_alignment_time = time.time() 164 | from hivision.creator.rotation_adjust import rotate_bound_4channels 165 | 166 | # 根据角度旋转原图和抠图 167 | b, g, r, a = cv2.split(ctx.matting_image) 168 | ctx.origin_image, ctx.matting_image, _, _, _, _ = rotate_bound_4channels( 169 | cv2.merge((b, g, r)), 170 | a, 171 | -1 * ctx.face["roll_angle"], 172 | ) 173 | 174 | # 旋转后再执行一遍人脸检测 175 | self.detection_handler(ctx) 176 | self.after_detect and self.after_detect(ctx) 177 | end_alignment_time = time.time() 178 | print(f"[3.1] Face Alignment Time: {end_alignment_time - start_alignment_time:.3f}s") 179 | 180 | # 4. ------------------图像调整------------------ 181 | print("[4] Start Image Post-Adjustment...") 182 | start_adjust_time = time.time() 183 | result_image_hd, result_image_standard, clothing_params, typography_params = ( 184 | adjust_photo(ctx) 185 | ) 186 | end_adjust_time = time.time() 187 | print(f"[4] Image Post-Adjustment Time: {end_adjust_time - start_adjust_time:.3f}s") 188 | 189 | # 5. ------------------返回结果------------------ 190 | ctx.result = Result( 191 | standard=result_image_standard, 192 | hd=result_image_hd, 193 | matting=ctx.matting_image, 194 | clothing_params=clothing_params, 195 | typography_params=typography_params, 196 | face=ctx.face, 197 | ) 198 | self.after_all and self.after_all(ctx) 199 | 200 | # 总的结束时间 201 | total_end_time = time.time() 202 | print(f"[Total] Total Time: {total_end_time - total_start_time:.3f}s") 203 | 204 | return ctx.result 205 | -------------------------------------------------------------------------------- /hivision/creator/choose_handler.py: -------------------------------------------------------------------------------- 1 | from hivision.creator.human_matting import * 2 | from hivision.creator.face_detector import * 3 | 4 | 5 | HUMAN_MATTING_MODELS = [ 6 | "modnet_photographic_portrait_matting", 7 | "birefnet-v1-lite", 8 | "hivision_modnet", 9 | "rmbg-1.4", 10 | ] 11 | 12 | FACE_DETECT_MODELS = ["face++ (联网Online API)", "mtcnn", "retinaface-resnet50"] 13 | 14 | 15 | def choose_handler(creator, matting_model_option=None, face_detect_option=None): 16 | if matting_model_option == "modnet_photographic_portrait_matting": 17 | creator.matting_handler = extract_human_modnet_photographic_portrait_matting 18 | elif matting_model_option == "mnn_hivision_modnet": 19 | creator.matting_handler = extract_human_mnn_modnet 20 | elif matting_model_option == "rmbg-1.4": 21 | creator.matting_handler = extract_human_rmbg 22 | elif matting_model_option == "birefnet-v1-lite": 23 | creator.matting_handler = extract_human_birefnet_lite 24 | else: 25 | creator.matting_handler = extract_human 26 | 27 | if ( 28 | face_detect_option == "face_plusplus" 29 | or face_detect_option == "face++ (联网Online API)" 30 | ): 31 | creator.detection_handler = detect_face_face_plusplus 32 | elif face_detect_option == "retinaface-resnet50": 33 | creator.detection_handler = detect_face_retinaface 34 | else: 35 | creator.detection_handler = detect_face_mtcnn 36 | -------------------------------------------------------------------------------- /hivision/creator/context.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding: utf-8 -*- 3 | r""" 4 | @DATE: 2024/9/5 19:20 5 | @File: context.py 6 | @IDE: pycharm 7 | @Description: 8 | 证件照创建上下文类，用于同步信息 9 | """ 10 | from typing import Optional, Callable, Tuple 11 | import numpy as np 12 | 13 | 14 | class Params: 15 | def __init__( 16 | self, 17 | size: Tuple[int, int] = (413, 295), 18 | change_bg_only: bool = False, 19 | crop_only: bool = False, 20 | head_measure_ratio: float = 0.2, 21 | head_height_ratio: float = 0.45, 22 | head_top_range: float = (0.12, 0.1), 23 | face: Tuple[int, int, int, int] = None, 24 | whitening_strength: int = 0, 25 | brightness_strength: int = 0, 26 | contrast_strength: int = 0, 27 | sharpen_strength: int = 0, 28 | saturation_strength: int = 0, 29 | face_alignment: bool = False, 30 | ): 31 | self.__size = size 32 | self.__change_bg_only = change_bg_only 33 | self.__crop_only = crop_only 34 | self.__head_measure_ratio = head_measure_ratio 35 | self.__head_height_ratio = head_height_ratio 36 | self.__head_top_range = head_top_range 37 | self.__face = face 38 | self.__whitening_strength = whitening_strength 39 | self.__brightness_strength = brightness_strength 40 | self.__contrast_strength = contrast_strength 41 | self.__sharpen_strength = sharpen_strength 42 | self.__saturation_strength = saturation_strength 43 | self.__face_alignment = face_alignment 44 | 45 | @property 46 | def size(self): 47 | return self.__size 48 | 49 | @property 50 | def change_bg_only(self): 51 | return self.__change_bg_only 52 | 53 | @property 54 | def head_measure_ratio(self): 55 | return self.__head_measure_ratio 56 | 57 | @property 58 | def head_height_ratio(self): 59 | return self.__head_height_ratio 60 | 61 | @property 62 | def head_top_range(self): 63 | return self.__head_top_range 64 | 65 | @property 66 | def crop_only(self): 67 | return self.__crop_only 68 | 69 | @property 70 | def face(self): 71 | return self.__face 72 | 73 | @property 74 | def whitening_strength(self): 75 | return self.__whitening_strength 76 | 77 | @property 78 | def brightness_strength(self): 79 | return self.__brightness_strength 80 | 81 | @property 82 | def contrast_strength(self): 83 | return self.__contrast_strength 84 | 85 | @property 86 | def sharpen_strength(self): 87 | return self.__sharpen_strength 88 | 89 | @property 90 | def saturation_strength(self): 91 | return self.__saturation_strength 92 | 93 | @property 94 | def face_alignment(self): 95 | return self.__face_alignment 96 | 97 | 98 | class Result: 99 | def __init__( 100 | self, 101 | standard: np.ndarray, 102 | hd: np.ndarray, 103 | matting: np.ndarray, 104 | clothing_params: Optional[dict], 105 | typography_params: Optional[dict], 106 | face: Optional[Tuple[int, int, int, int, float]], 107 | ): 108 | self.standard = standard 109 | self.hd = hd 110 | self.matting = matting 111 | self.clothing_params = clothing_params 112 | """ 113 | 服装参数，仅换底时为 None 114 | """ 115 | self.typography_params = typography_params 116 | """ 117 | 排版参数，仅换底时为 None 118 | """ 119 | self.face = face 120 | 121 | def __iter__(self): 122 | return iter( 123 | [ 124 | self.standard, 125 | self.hd, 126 | self.matting, 127 | self.clothing_params, 128 | self.typography_params, 129 | self.face, 130 | ] 131 | ) 132 | 133 | 134 | class Context: 135 | def __init__(self, params: Params): 136 | self.params: Params = params 137 | """ 138 | 证件照处理参数 139 | """ 140 | self.origin_image: Optional[np.ndarray] = None 141 | """ 142 | 输入的原始图像，处理时会进行resize，长宽不一定等于输入图像 143 | """ 144 | self.processing_image: Optional[np.ndarray] = None 145 | """ 146 | 当前正在处理的图像 147 | """ 148 | self.matting_image: Optional[np.ndarray] = None 149 | """ 150 | 人像抠图结果 151 | """ 152 | self.face: dict = dict(rectangle=None, roll_angle=None) 153 | """ 154 | 人脸检测结果，大于一个人脸时已在上层抛出异常 155 | rectangle: 人脸矩形框，包含 x1, y1, width, height 的坐标, x1, y1 为左上角坐标, width, height 为矩形框的宽度和高度 156 | roll_angle: 人脸偏转角度，以眼睛为标准，计算的人脸偏转角度，用于人脸矫正 157 | """ 158 | self.result: Optional[Result] = None 159 | """ 160 | 证件照处理结果 161 | """ 162 | self.align_info: Optional[dict] = None 163 | """ 164 | 人脸矫正信息，仅当 align_face 为 True 时存在 165 | """ 166 | 167 | 168 | ContextHandler = Optional[Callable[[Context], None]] 169 | -------------------------------------------------------------------------------- /hivision/creator/face_detector.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding: utf-8 -*- 3 | r""" 4 | @DATE: 2024/9/5 19:32 5 | @File: face_detector.py 6 | @IDE: pycharm 7 | @Description: 8 | 人脸检测器 9 | """ 10 | try: 11 | from mtcnnruntime import MTCNN 12 | except ImportError: 13 | raise ImportError( 14 | "Please install mtcnn-runtime by running `pip install mtcnn-runtime`" 15 | ) 16 | from .context import Context 17 | from hivision.error import FaceError, APIError 18 | from hivision.utils import resize_image_to_kb_base64 19 | from hivision.creator.retinaface import retinaface_detect_faces 20 | import requests 21 | import cv2 22 | import os 23 | import numpy as np 24 | 25 | 26 | mtcnn = None 27 | base_dir = os.path.dirname(os.path.abspath(__file__)) 28 | RETINAFCE_SESS = None 29 | 30 | 31 | def detect_face_mtcnn(ctx: Context, scale: int = 2): 32 | """ 33 | 基于MTCNN模型的人脸检测处理器，只进行人脸数量的检测 34 | :param ctx: 上下文，此时已获取到原始图和抠图结果，但是我们只需要原始图 35 | :param scale: 最大边长缩放比例，原图:缩放图 = 1:scale 36 | :raise FaceError: 人脸检测错误，多个人脸或者没有人脸 37 | """ 38 | global mtcnn 39 | if mtcnn is None: 40 | mtcnn = MTCNN() 41 | image = cv2.resize( 42 | ctx.origin_image, 43 | (ctx.origin_image.shape[1] // scale, ctx.origin_image.shape[0] // scale), 44 | interpolation=cv2.INTER_AREA, 45 | ) 46 | # landmarks 是 5 个关键点，分别是左眼、右眼、鼻子、左嘴角、右嘴角， 47 | faces, landmarks = mtcnn.detect(image, thresholds=[0.8, 0.8, 0.8]) 48 | 49 | # print(len(faces)) 50 | if len(faces) != 1: 51 | # 保险措施，如果检测到多个人脸或者没有人脸，用原图再检测一次 52 | faces, landmarks = mtcnn.detect(ctx.origin_image) 53 | else: 54 | # 如果只有一个人脸，将人脸坐标放大 55 | for item, param in enumerate(faces[0]): 56 | faces[0][item] = param * 2 57 | if len(faces) != 1: 58 | raise FaceError("Expected 1 face, but got {}".format(len(faces)), len(faces)) 59 | 60 | # 计算人脸坐标 61 | left = faces[0][0] 62 | top = faces[0][1] 63 | width = faces[0][2] - left + 1 64 | height = faces[0][3] - top + 1 65 | ctx.face["rectangle"] = (left, top, width, height) 66 | 67 | # 根据landmarks计算人脸偏转角度，以眼睛为标准，计算的人脸偏转角度，用于人脸矫正 68 | # 示例landmarks [106.37181 150.77415 127.21012 108.369156 144.61522 105.24723 107.45625 133.62355 151.24269 153.34407 ] 69 | landmarks = landmarks[0] 70 | left_eye = np.array([landmarks[0], landmarks[5]]) 71 | right_eye = np.array([landmarks[1], landmarks[6]]) 72 | dy = right_eye[1] - left_eye[1] 73 | dx = right_eye[0] - left_eye[0] 74 | roll_angle = np.degrees(np.arctan2(dy, dx)) 75 | 76 | ctx.face["roll_angle"] = roll_angle 77 | 78 | 79 | def detect_face_face_plusplus(ctx: Context): 80 | """ 81 | 基于Face++ API接口的人脸检测处理器，只进行人脸数量的检测 82 | :param ctx: 上下文，此时已获取到原始图和抠图结果，但是我们只需要原始图 83 | :param scale: 最大边长缩放比例，原图:缩放图 = 1:scale 84 | :raise FaceError: 人脸检测错误，多个人脸或者没有人脸 85 | """ 86 | url = "https://api-cn.faceplusplus.com/facepp/v3/detect" 87 | api_key = os.getenv("FACE_PLUS_API_KEY") 88 | api_secret = os.getenv("FACE_PLUS_API_SECRET") 89 | 90 | print("调用了face++") 91 | 92 | image = ctx.origin_image 93 | # 将图片转为 base64, 且不大于2MB（Face++ API接口限制） 94 | image_base64 = resize_image_to_kb_base64(image, 2000, mode="max") 95 | 96 | files = { 97 | "api_key": (None, api_key), 98 | "api_secret": (None, api_secret), 99 | "image_base64": (None, image_base64), 100 | "return_landmark": (None, "1"), 101 | "return_attributes": (None, "headpose"), 102 | } 103 | 104 | # 发送 POST 请求 105 | response = requests.post(url, files=files) 106 | 107 | # 获取响应状态码 108 | status_code = response.status_code 109 | response_json = response.json() 110 | 111 | if status_code == 200: 112 | face_num = response_json["face_num"] 113 | if face_num == 1: 114 | face_rectangle = response_json["faces"][0]["face_rectangle"] 115 | 116 | # 获取人脸关键点 117 | # landmarks = response_json["faces"][0]["landmark"] 118 | # print("face++ landmarks", landmarks) 119 | 120 | # headpose 是一个字典，包含俯仰角（pitch）、偏航角（yaw）和滚转角（roll） 121 | # headpose示例 {'pitch_angle': 6.997899, 'roll_angle': 1.8011835, 'yaw_angle': 5.043002} 122 | headpose = response_json["faces"][0]["attributes"]["headpose"] 123 | # 以眼睛为标准，计算的人脸偏转角度，用于人脸矫正 124 | roll_angle = headpose["roll_angle"] / 2 125 | 126 | ctx.face["rectangle"] = ( 127 | face_rectangle["left"], 128 | face_rectangle["top"], 129 | face_rectangle["width"], 130 | face_rectangle["height"], 131 | ) 132 | ctx.face["roll_angle"] = roll_angle 133 | else: 134 | raise FaceError( 135 | "Expected 1 face, but got {}".format(face_num), len(face_num) 136 | ) 137 | 138 | elif status_code == 401: 139 | raise APIError( 140 | f"Face++ Status code {status_code} Authentication error: API key and secret do not match.", 141 | status_code, 142 | ) 143 | 144 | elif status_code == 403: 145 | reason = response_json.get("error_message", "Unknown authorization error.") 146 | raise APIError( 147 | f"Authorization error: {reason}", 148 | status_code, 149 | ) 150 | 151 | elif status_code == 400: 152 | error_message = response_json.get("error_message", "Bad request.") 153 | raise APIError( 154 | f"Bad request error: {error_message}", 155 | status_code, 156 | ) 157 | 158 | elif status_code == 413: 159 | raise APIError( 160 | f"Face++ Status code {status_code} Request entity too large: The image exceeds the 2MB limit.", 161 | status_code, 162 | ) 163 | 164 | 165 | def detect_face_retinaface(ctx: Context): 166 | """ 167 | 基于RetinaFace模型的人脸检测处理器，只进行人脸数量的检测 168 | :param ctx: 上下文，此时已获取到原始图和抠图结果，但是我们只需要原始图 169 | :raise FaceError: 人脸检测错误，多个人脸或者没有人脸 170 | """ 171 | from time import time 172 | 173 | global RETINAFCE_SESS 174 | 175 | if RETINAFCE_SESS is None: 176 | # 计算用时 177 | tic = time() 178 | faces_dets, sess = retinaface_detect_faces( 179 | ctx.origin_image, 180 | os.path.join(base_dir, "retinaface/weights/retinaface-resnet50.onnx"), 181 | sess=None, 182 | ) 183 | RETINAFCE_SESS = sess 184 | else: 185 | tic = time() 186 | faces_dets, _ = retinaface_detect_faces( 187 | ctx.origin_image, 188 | os.path.join(base_dir, "retinaface/weights/retinaface-resnet50.onnx"), 189 | sess=RETINAFCE_SESS, 190 | ) 191 | 192 | faces_num = len(faces_dets) 193 | faces_landmarks = [] 194 | for face_det in faces_dets: 195 | faces_landmarks.append(face_det[5:]) 196 | 197 | if faces_num != 1: 198 | raise FaceError("Expected 1 face, but got {}".format(faces_num), faces_num) 199 | face_det = faces_dets[0] 200 | ctx.face["rectangle"] = ( 201 | face_det[0], 202 | face_det[1], 203 | face_det[2] - face_det[0] + 1, 204 | face_det[3] - face_det[1] + 1, 205 | ) 206 | 207 | # 计算roll_angle 208 | face_landmarks = faces_landmarks[0] 209 | # print("face_landmarks", face_landmarks) 210 | left_eye = np.array([face_landmarks[0], face_landmarks[1]]) 211 | right_eye = np.array([face_landmarks[2], face_landmarks[3]]) 212 | dy = right_eye[1] - left_eye[1] 213 | dx = right_eye[0] - left_eye[0] 214 | roll_angle = np.degrees(np.arctan2(dy, dx)) 215 | ctx.face["roll_angle"] = roll_angle 216 | 217 | # 如果RUN_MODE不是野兽模式，则释放模型 218 | if os.getenv("RUN_MODE") == "beast": 219 | RETINAFCE_SESS = None -------------------------------------------------------------------------------- /hivision/creator/human_matting.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding: utf-8 -*- 3 | r""" 4 | @DATE: 2024/9/5 21:21 5 | @File: human_matting.py 6 | @IDE: pycharm 7 | @Description: 8 | 人像抠图 9 | """ 10 | import numpy as np 11 | from PIL import Image 12 | import onnxruntime 13 | from .tensor2numpy import NNormalize, NTo_Tensor, NUnsqueeze 14 | from .context import Context 15 | import cv2 16 | import os 17 | from time import time 18 | 19 | 20 | WEIGHTS = { 21 | "hivision_modnet": os.path.join( 22 | os.path.dirname(__file__), "weights", "hivision_modnet.onnx" 23 | ), 24 | "modnet_photographic_portrait_matting": os.path.join( 25 | os.path.dirname(__file__), 26 | "weights", 27 | "modnet_photographic_portrait_matting.onnx", 28 | ), 29 | "mnn_hivision_modnet": os.path.join( 30 | os.path.dirname(__file__), 31 | "weights", 32 | "mnn_hivision_modnet.mnn", 33 | ), 34 | "rmbg-1.4": os.path.join(os.path.dirname(__file__), "weights", "rmbg-1.4.onnx"), 35 | "birefnet-v1-lite": os.path.join( 36 | os.path.dirname(__file__), "weights", "birefnet-v1-lite.onnx" 37 | ), 38 | } 39 | 40 | ONNX_DEVICE = onnxruntime.get_device() 41 | ONNX_PROVIDER = ( 42 | "CUDAExecutionProvider" if ONNX_DEVICE == "GPU" else "CPUExecutionProvider" 43 | ) 44 | 45 | HIVISION_MODNET_SESS = None 46 | MODNET_PHOTOGRAPHIC_PORTRAIT_MATTING_SESS = None 47 | RMBG_SESS = None 48 | BIREFNET_V1_LITE_SESS = None 49 | 50 | 51 | def load_onnx_model(checkpoint_path, set_cpu=False): 52 | providers = ( 53 | ["CUDAExecutionProvider", "CPUExecutionProvider"] 54 | if ONNX_PROVIDER == "CUDAExecutionProvider" 55 | else ["CPUExecutionProvider"] 56 | ) 57 | 58 | if set_cpu: 59 | sess = onnxruntime.InferenceSession( 60 | checkpoint_path, providers=["CPUExecutionProvider"] 61 | ) 62 | else: 63 | try: 64 | sess = onnxruntime.InferenceSession(checkpoint_path, providers=providers) 65 | except Exception as e: 66 | if ONNX_DEVICE == "CUDAExecutionProvider": 67 | print(f"Failed to load model with CUDAExecutionProvider: {e}") 68 | print("Falling back to CPUExecutionProvider") 69 | # 尝试使用CPU加载模型 70 | sess = onnxruntime.InferenceSession( 71 | checkpoint_path, providers=["CPUExecutionProvider"] 72 | ) 73 | else: 74 | raise e # 如果是CPU执行失败，重新抛出异常 75 | 76 | return sess 77 | 78 | 79 | def extract_human(ctx: Context): 80 | """ 81 | 人像抠图 82 | :param ctx: 上下文 83 | """ 84 | # 抠图 85 | matting_image = get_modnet_matting(ctx.processing_image, WEIGHTS["hivision_modnet"]) 86 | # 修复抠图 87 | ctx.processing_image = hollow_out_fix(matting_image) 88 | ctx.matting_image = ctx.processing_image.copy() 89 | 90 | 91 | def extract_human_modnet_photographic_portrait_matting(ctx: Context): 92 | """ 93 | 人像抠图 94 | :param ctx: 上下文 95 | """ 96 | # 抠图 97 | matting_image = get_modnet_matting_photographic_portrait_matting( 98 | ctx.processing_image, WEIGHTS["modnet_photographic_portrait_matting"] 99 | ) 100 | # 修复抠图 101 | ctx.processing_image = matting_image 102 | ctx.matting_image = ctx.processing_image.copy() 103 | 104 | 105 | def extract_human_mnn_modnet(ctx: Context): 106 | matting_image = get_mnn_modnet_matting( 107 | ctx.processing_image, WEIGHTS["mnn_hivision_modnet"] 108 | ) 109 | ctx.processing_image = hollow_out_fix(matting_image) 110 | ctx.matting_image = ctx.processing_image.copy() 111 | 112 | 113 | def extract_human_rmbg(ctx: Context): 114 | matting_image = get_rmbg_matting(ctx.processing_image, WEIGHTS["rmbg-1.4"]) 115 | ctx.processing_image = matting_image 116 | ctx.matting_image = ctx.processing_image.copy() 117 | 118 | 119 | # def extract_human_birefnet_portrait(ctx: Context): 120 | # matting_image = get_birefnet_portrait_matting( 121 | # ctx.processing_image, WEIGHTS["birefnet-portrait"] 122 | # ) 123 | # ctx.processing_image = matting_image 124 | # ctx.matting_image = ctx.processing_image.copy() 125 | 126 | 127 | def extract_human_birefnet_lite(ctx: Context): 128 | matting_image = get_birefnet_portrait_matting( 129 | ctx.processing_image, WEIGHTS["birefnet-v1-lite"] 130 | ) 131 | ctx.processing_image = matting_image 132 | ctx.matting_image = ctx.processing_image.copy() 133 | 134 | 135 | def hollow_out_fix(src: np.ndarray) -> np.ndarray: 136 | """ 137 | 修补抠图区域，作为抠图模型精度不够的补充 138 | :param src: 139 | :return: 140 | """ 141 | b, g, r, a = cv2.split(src) 142 | src_bgr = cv2.merge((b, g, r)) 143 | # -----------padding---------- # 144 | add_area = np.zeros((10, a.shape[1]), np.uint8) 145 | a = np.vstack((add_area, a, add_area)) 146 | add_area = np.zeros((a.shape[0], 10), np.uint8) 147 | a = np.hstack((add_area, a, add_area)) 148 | # -------------end------------ # 149 | _, a_threshold = cv2.threshold(a, 127, 255, 0) 150 | a_erode = cv2.erode( 151 | a_threshold, 152 | kernel=cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5)), 153 | iterations=3, 154 | ) 155 | contours, hierarchy = cv2.findContours( 156 | a_erode, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE 157 | ) 158 | contours = [x for x in contours] 159 | # contours = np.squeeze(contours) 160 | contours.sort(key=lambda c: cv2.contourArea(c), reverse=True) 161 | a_contour = cv2.drawContours(np.zeros(a.shape, np.uint8), contours[0], -1, 255, 2) 162 | # a_base = a_contour[1:-1, 1:-1] 163 | h, w = a.shape[:2] 164 | mask = np.zeros( 165 | [h + 2, w + 2], np.uint8 166 | ) # mask 必须行和列都加 2，且必须为 uint8 单通道阵列 167 | cv2.floodFill(a_contour, mask=mask, seedPoint=(0, 0), newVal=255) 168 | a = cv2.add(a, 255 - a_contour) 169 | return cv2.merge((src_bgr, a[10:-10, 10:-10])) 170 | 171 | 172 | def image2bgr(input_image): 173 | if len(input_image.shape) == 2: 174 | input_image = input_image[:, :, None] 175 | if input_image.shape[2] == 1: 176 | result_image = np.repeat(input_image, 3, axis=2) 177 | elif input_image.shape[2] == 4: 178 | result_image = input_image[:, :, 0:3] 179 | else: 180 | result_image = input_image 181 | 182 | return result_image 183 | 184 | 185 | def read_modnet_image(input_image, ref_size=512): 186 | im = Image.fromarray(np.uint8(input_image)) 187 | width, length = im.size[0], im.size[1] 188 | im = np.asarray(im) 189 | im = image2bgr(im) 190 | im = cv2.resize(im, (ref_size, ref_size), interpolation=cv2.INTER_AREA) 191 | im = NNormalize(im, mean=np.array([0.5, 0.5, 0.5]), std=np.array([0.5, 0.5, 0.5])) 192 | im = NUnsqueeze(NTo_Tensor(im)) 193 | 194 | return im, width, length 195 | 196 | 197 | def get_modnet_matting(input_image, checkpoint_path, ref_size=512): 198 | global HIVISION_MODNET_SESS 199 | 200 | if not os.path.exists(checkpoint_path): 201 | print(f"Checkpoint file not found: {checkpoint_path}") 202 | return None 203 | 204 | # 如果RUN_MODE不是野兽模式，则不加载模型 205 | if HIVISION_MODNET_SESS is None: 206 | HIVISION_MODNET_SESS = load_onnx_model(checkpoint_path, set_cpu=True) 207 | 208 | input_name = HIVISION_MODNET_SESS.get_inputs()[0].name 209 | output_name = HIVISION_MODNET_SESS.get_outputs()[0].name 210 | 211 | im, width, length = read_modnet_image(input_image=input_image, ref_size=ref_size) 212 | 213 | matte = HIVISION_MODNET_SESS.run([output_name], {input_name: im}) 214 | matte = (matte[0] * 255).astype("uint8") 215 | matte = np.squeeze(matte) 216 | mask = cv2.resize(matte, (width, length), interpolation=cv2.INTER_AREA) 217 | b, g, r = cv2.split(np.uint8(input_image)) 218 | 219 | output_image = cv2.merge((b, g, r, mask)) 220 | 221 | # 如果RUN_MODE不是野兽模式，则释放模型 222 | if os.getenv("RUN_MODE") != "beast": 223 | HIVISION_MODNET_SESS = None 224 | 225 | return output_image 226 | 227 | 228 | def get_modnet_matting_photographic_portrait_matting( 229 | input_image, checkpoint_path, ref_size=512 230 | ): 231 | global MODNET_PHOTOGRAPHIC_PORTRAIT_MATTING_SESS 232 | 233 | if not os.path.exists(checkpoint_path): 234 | print(f"Checkpoint file not found: {checkpoint_path}") 235 | return None 236 | 237 | # 如果RUN_MODE不是野兽模式，则不加载模型 238 | if MODNET_PHOTOGRAPHIC_PORTRAIT_MATTING_SESS is None: 239 | MODNET_PHOTOGRAPHIC_PORTRAIT_MATTING_SESS = load_onnx_model( 240 | checkpoint_path, set_cpu=True 241 | ) 242 | 243 | input_name = MODNET_PHOTOGRAPHIC_PORTRAIT_MATTING_SESS.get_inputs()[0].name 244 | output_name = MODNET_PHOTOGRAPHIC_PORTRAIT_MATTING_SESS.get_outputs()[0].name 245 | 246 | im, width, length = read_modnet_image(input_image=input_image, ref_size=ref_size) 247 | 248 | matte = MODNET_PHOTOGRAPHIC_PORTRAIT_MATTING_SESS.run( 249 | [output_name], {input_name: im} 250 | ) 251 | matte = (matte[0] * 255).astype("uint8") 252 | matte = np.squeeze(matte) 253 | mask = cv2.resize(matte, (width, length), interpolation=cv2.INTER_AREA) 254 | b, g, r = cv2.split(np.uint8(input_image)) 255 | 256 | output_image = cv2.merge((b, g, r, mask)) 257 | 258 | # 如果RUN_MODE不是野兽模式，则释放模型 259 | if os.getenv("RUN_MODE") != "beast": 260 | MODNET_PHOTOGRAPHIC_PORTRAIT_MATTING_SESS = None 261 | 262 | return output_image 263 | 264 | 265 | def get_rmbg_matting(input_image: np.ndarray, checkpoint_path, ref_size=1024): 266 | global RMBG_SESS 267 | 268 | if not os.path.exists(checkpoint_path): 269 | print(f"Checkpoint file not found: {checkpoint_path}") 270 | return None 271 | 272 | def resize_rmbg_image(image): 273 | image = image.convert("RGB") 274 | model_input_size = (ref_size, ref_size) 275 | image = image.resize(model_input_size, Image.BILINEAR) 276 | return image 277 | 278 | if RMBG_SESS is None: 279 | RMBG_SESS = load_onnx_model(checkpoint_path, set_cpu=True) 280 | 281 | orig_image = Image.fromarray(input_image) 282 | image = resize_rmbg_image(orig_image) 283 | im_np = np.array(image).astype(np.float32) 284 | im_np = im_np.transpose(2, 0, 1) # Change to CxHxW format 285 | im_np = np.expand_dims(im_np, axis=0) # Add batch dimension 286 | im_np = im_np / 255.0 # Normalize to [0, 1] 287 | im_np = (im_np - 0.5) / 0.5 # Normalize to [-1, 1] 288 | 289 | # Inference 290 | result = RMBG_SESS.run(None, {RMBG_SESS.get_inputs()[0].name: im_np})[0] 291 | 292 | # Post process 293 | result = np.squeeze(result) 294 | ma = np.max(result) 295 | mi = np.min(result) 296 | result = (result - mi) / (ma - mi) # Normalize to [0, 1] 297 | 298 | # Convert to PIL image 299 | im_array = (result * 255).astype(np.uint8) 300 | pil_im = Image.fromarray( 301 | im_array, mode="L" 302 | ) # Ensure mask is single channel (L mode) 303 | 304 | # Resize the mask to match the original image size 305 | pil_im = pil_im.resize(orig_image.size, Image.BILINEAR) 306 | 307 | # Paste the mask on the original image 308 | new_im = Image.new("RGBA", orig_image.size, (0, 0, 0, 0)) 309 | new_im.paste(orig_image, mask=pil_im) 310 | 311 | # 如果RUN_MODE不是野兽模式，则释放模型 312 | if os.getenv("RUN_MODE") != "beast": 313 | RMBG_SESS = None 314 | 315 | return np.array(new_im) 316 | 317 | 318 | def get_mnn_modnet_matting(input_image, checkpoint_path, ref_size=512): 319 | if not os.path.exists(checkpoint_path): 320 | print(f"Checkpoint file not found: {checkpoint_path}") 321 | return None 322 | 323 | try: 324 | import MNN.expr as expr 325 | import MNN.nn as nn 326 | except ImportError as e: 327 | raise ImportError( 328 | "The MNN module is not installed or there was an import error. Please ensure that the MNN library is installed by using the command 'pip install mnn'." 329 | ) from e 330 | 331 | config = {} 332 | config["precision"] = "low" # 当硬件支持（armv8.2）时使用fp16推理 333 | config["backend"] = 0 # CPU 334 | config["numThread"] = 4 # 线程数 335 | im, width, length = read_modnet_image(input_image, ref_size=512) 336 | rt = nn.create_runtime_manager((config,)) 337 | net = nn.load_module_from_file( 338 | checkpoint_path, ["input1"], ["output1"], runtime_manager=rt 339 | ) 340 | input_var = expr.convert(im, expr.NCHW) 341 | output_var = net.forward(input_var) 342 | matte = expr.convert(output_var, expr.NCHW) 343 | matte = matte.read() # var转换为np 344 | matte = (matte * 255).astype("uint8") 345 | matte = np.squeeze(matte) 346 | mask = cv2.resize(matte, (width, length), interpolation=cv2.INTER_AREA) 347 | b, g, r = cv2.split(np.uint8(input_image)) 348 | 349 | output_image = cv2.merge((b, g, r, mask)) 350 | 351 | return output_image 352 | 353 | 354 | def get_birefnet_portrait_matting(input_image, checkpoint_path, ref_size=512): 355 | global BIREFNET_V1_LITE_SESS 356 | 357 | if not os.path.exists(checkpoint_path): 358 | print(f"Checkpoint file not found: {checkpoint_path}") 359 | return None 360 | 361 | def transform_image(image): 362 | image = image.resize((1024, 1024)) # Resize to 1024x1024 363 | image = ( 364 | np.array(image, dtype=np.float32) / 255.0 365 | ) # Convert to numpy array and normalize to [0, 1] 366 | image = (image - [0.485, 0.456, 0.406]) / [0.229, 0.224, 0.225] # Normalize 367 | image = np.transpose(image, (2, 0, 1)) # Change from (H, W, C) to (C, H, W) 368 | image = np.expand_dims(image, axis=0) # Add batch dimension 369 | return image.astype(np.float32) # Ensure the output is float32 370 | 371 | orig_image = Image.fromarray(input_image) 372 | input_images = transform_image( 373 | orig_image 374 | ) # This will already have the correct shape 375 | 376 | # 记录加载onnx模型的开始时间 377 | load_start_time = time() 378 | 379 | # 如果RUN_MODE不是野兽模式，则不加载模型 380 | if BIREFNET_V1_LITE_SESS is None: 381 | # print("首次加载birefnet-v1-lite模型...") 382 | if ONNX_DEVICE == "GPU": 383 | print("onnxruntime-gpu已安装，尝试使用CUDA加载模型") 384 | try: 385 | import torch 386 | except ImportError: 387 | print( 388 | "torch未安装，尝试直接使用onnxruntime-gpu加载模型，这需要配置好CUDA和cuDNN" 389 | ) 390 | BIREFNET_V1_LITE_SESS = load_onnx_model(checkpoint_path) 391 | else: 392 | BIREFNET_V1_LITE_SESS = load_onnx_model(checkpoint_path, set_cpu=True) 393 | 394 | # 记录加载onnx模型的结束时间 395 | load_end_time = time() 396 | 397 | # 打印加载onnx模型所花的时间 398 | print(f"Loading ONNX model took {load_end_time - load_start_time:.4f} seconds") 399 | 400 | input_name = BIREFNET_V1_LITE_SESS.get_inputs()[0].name 401 | print(onnxruntime.get_device(), BIREFNET_V1_LITE_SESS.get_providers()) 402 | 403 | time_st = time() 404 | pred_onnx = BIREFNET_V1_LITE_SESS.run(None, {input_name: input_images})[ 405 | -1 406 | ] # Use float32 input 407 | pred_onnx = np.squeeze(pred_onnx) # Use numpy to squeeze 408 | result = 1 / (1 + np.exp(-pred_onnx)) # Sigmoid function using numpy 409 | print(f"Inference time: {time() - time_st:.4f} seconds") 410 | 411 | # Convert to PIL image 412 | im_array = (result * 255).astype(np.uint8) 413 | pil_im = Image.fromarray( 414 | im_array, mode="L" 415 | ) # Ensure mask is single channel (L mode) 416 | 417 | # Resize the mask to match the original image size 418 | pil_im = pil_im.resize(orig_image.size, Image.BILINEAR) 419 | 420 | # Paste the mask on the original image 421 | new_im = Image.new("RGBA", orig_image.size, (0, 0, 0, 0)) 422 | new_im.paste(orig_image, mask=pil_im) 423 | 424 | # 如果RUN_MODE不是野兽模式，则释放模型 425 | if os.getenv("RUN_MODE") != "beast": 426 | BIREFNET_V1_LITE_SESS = None 427 | 428 | return np.array(new_im) 429 | -------------------------------------------------------------------------------- /hivision/creator/layout_calculator.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding: utf-8 -*- 3 | r""" 4 | @DATE: 2024/9/5 21:35 5 | @File: layout_calculator.py 6 | @IDE: pycharm 7 | @Description: 8 | 布局计算器 9 | """ 10 | 11 | import cv2.detail 12 | import numpy as np 13 | 14 | 15 | def judge_layout( 16 | input_width, 17 | input_height, 18 | PHOTO_INTERVAL_W, 19 | PHOTO_INTERVAL_H, 20 | LIMIT_BLOCK_W, 21 | LIMIT_BLOCK_H, 22 | ): 23 | centerBlockHeight_1, centerBlockWidth_1 = ( 24 | input_height, 25 | input_width, 26 | ) # 由证件照们组成的一个中心区块（1 代表不转置排列） 27 | centerBlockHeight_2, centerBlockWidth_2 = ( 28 | input_width, 29 | input_height, 30 | ) # 由证件照们组成的一个中心区块（2 代表转置排列） 31 | 32 | # 1.不转置排列的情况下： 33 | layout_col_no_transpose = 0 # 行 34 | layout_row_no_transpose = 0 # 列 35 | for i in range(1, 4): 36 | centerBlockHeight_temp = input_height * i + PHOTO_INTERVAL_H * (i - 1) 37 | if centerBlockHeight_temp < LIMIT_BLOCK_H: 38 | centerBlockHeight_1 = centerBlockHeight_temp 39 | layout_row_no_transpose = i 40 | else: 41 | break 42 | for j in range(1, 9): 43 | centerBlockWidth_temp = input_width * j + PHOTO_INTERVAL_W * (j - 1) 44 | if centerBlockWidth_temp < LIMIT_BLOCK_W: 45 | centerBlockWidth_1 = centerBlockWidth_temp 46 | layout_col_no_transpose = j 47 | else: 48 | break 49 | layout_number_no_transpose = layout_row_no_transpose * layout_col_no_transpose 50 | 51 | # 2.转置排列的情况下： 52 | layout_col_transpose = 0 # 行 53 | layout_row_transpose = 0 # 列 54 | for i in range(1, 4): 55 | centerBlockHeight_temp = input_width * i + PHOTO_INTERVAL_H * (i - 1) 56 | if centerBlockHeight_temp < LIMIT_BLOCK_H: 57 | centerBlockHeight_2 = centerBlockHeight_temp 58 | layout_row_transpose = i 59 | else: 60 | break 61 | for j in range(1, 9): 62 | centerBlockWidth_temp = input_height * j + PHOTO_INTERVAL_W * (j - 1) 63 | if centerBlockWidth_temp < LIMIT_BLOCK_W: 64 | centerBlockWidth_2 = centerBlockWidth_temp 65 | layout_col_transpose = j 66 | else: 67 | break 68 | layout_number_transpose = layout_row_transpose * layout_col_transpose 69 | 70 | if layout_number_transpose > layout_number_no_transpose: 71 | layout_mode = (layout_col_transpose, layout_row_transpose, 2) 72 | return layout_mode, centerBlockWidth_2, centerBlockHeight_2 73 | else: 74 | layout_mode = (layout_col_no_transpose, layout_row_no_transpose, 1) 75 | return layout_mode, centerBlockWidth_1, centerBlockHeight_1 76 | 77 | 78 | def generate_layout_array(input_height, input_width, LAYOUT_WIDTH=1795, LAYOUT_HEIGHT=1205): 79 | # 1.基础参数表 80 | PHOTO_INTERVAL_H = 30 # 证件照与证件照之间的垂直距离 81 | PHOTO_INTERVAL_W = 30 # 证件照与证件照之间的水平距离 82 | SIDES_INTERVAL_H = 50 # 证件照与画布边缘的垂直距离 83 | SIDES_INTERVAL_W = 70 # 证件照与画布边缘的水平距离 84 | LIMIT_BLOCK_W = LAYOUT_WIDTH - 2 * SIDES_INTERVAL_W 85 | LIMIT_BLOCK_H = LAYOUT_HEIGHT - 2 * SIDES_INTERVAL_H 86 | 87 | # 2.创建一个 1180x1746 的空白画布 88 | white_background = np.zeros([LAYOUT_HEIGHT, LAYOUT_WIDTH, 3], np.uint8) 89 | white_background.fill(255) 90 | 91 | # 3.计算照片的 layout（列、行、横竖朝向）,证件照组成的中心区块的分辨率 92 | layout_mode, centerBlockWidth, centerBlockHeight = judge_layout( 93 | input_width, 94 | input_height, 95 | PHOTO_INTERVAL_W, 96 | PHOTO_INTERVAL_H, 97 | LIMIT_BLOCK_W, 98 | LIMIT_BLOCK_H, 99 | ) 100 | # 4.开始排列组合 101 | x11 = (LAYOUT_WIDTH - centerBlockWidth) // 2 102 | y11 = (LAYOUT_HEIGHT - centerBlockHeight) // 2 103 | typography_arr = [] 104 | typography_rotate = False 105 | if layout_mode[2] == 2: 106 | input_height, input_width = input_width, input_height 107 | typography_rotate = True 108 | 109 | for j in range(layout_mode[1]): 110 | for i in range(layout_mode[0]): 111 | xi = x11 + i * input_width + i * PHOTO_INTERVAL_W 112 | yi = y11 + j * input_height + j * PHOTO_INTERVAL_H 113 | typography_arr.append([xi, yi]) 114 | 115 | return typography_arr, typography_rotate 116 | 117 | 118 | def generate_layout_image( 119 | input_image, typography_arr, typography_rotate, width=295, height=413, 120 | crop_line:bool=False, 121 | LAYOUT_WIDTH=1795, 122 | LAYOUT_HEIGHT=1205, 123 | ): 124 | 125 | # 创建一个白色背景的空白画布 126 | white_background = np.zeros([LAYOUT_HEIGHT, LAYOUT_WIDTH, 3], np.uint8) 127 | white_background.fill(255) 128 | 129 | # 如果输入图像的高度不等于指定高度，则调整图像大小 130 | if input_image.shape[0] != height: 131 | input_image = cv2.resize(input_image, (width, height)) 132 | 133 | # 如果需要旋转排版，则对图像进行转置和垂直镜像 134 | if typography_rotate: 135 | input_image = cv2.transpose(input_image) 136 | input_image = cv2.flip(input_image, 0) # 0 表示垂直镜像 137 | 138 | # 交换高度和宽度 139 | height, width = width, height 140 | 141 | # 将图像按照排版数组中的位置放置到白色背景上 142 | for arr in typography_arr: 143 | locate_x, locate_y = arr[0], arr[1] 144 | white_background[locate_y : locate_y + height, locate_x : locate_x + width] = ( 145 | input_image 146 | ) 147 | 148 | if crop_line: 149 | # 添加裁剪线 150 | line_color = (200, 200, 200) # 浅灰色 151 | line_thickness = 1 152 | 153 | # 初始化裁剪线位置列表 154 | vertical_lines = [] 155 | horizontal_lines = [] 156 | 157 | # 根据排版数组添加裁剪线 158 | for arr in typography_arr: 159 | x, y = arr[0], arr[1] 160 | if x not in vertical_lines: 161 | vertical_lines.append(x) 162 | if x + width not in vertical_lines: 163 | vertical_lines.append(x + width) 164 | if y not in horizontal_lines: 165 | horizontal_lines.append(y) 166 | if y + height not in horizontal_lines: 167 | horizontal_lines.append(y + height) 168 | 169 | # 绘制垂直裁剪线 170 | for x in vertical_lines: 171 | cv2.line(white_background, (x, 0), (x, LAYOUT_HEIGHT), line_color, line_thickness) 172 | 173 | # 绘制水平裁剪线 174 | for y in horizontal_lines: 175 | cv2.line(white_background, (0, y), (LAYOUT_WIDTH, y), line_color, line_thickness) 176 | 177 | # 返回排版后的图像 178 | return white_background 179 | -------------------------------------------------------------------------------- /hivision/creator/move_image.py: -------------------------------------------------------------------------------- 1 | """ 2 | 有一些 png 图像下部也会有一些透明的区域，使得图像无法对其底部边框 3 | 本程序实现移动图像，使其下部与 png 图像实际大小相对齐 4 | """ 5 | import os 6 | import cv2 7 | import numpy as np 8 | from hivisionai.hycv.utils import get_box_pro 9 | 10 | path_pre = os.path.join(os.getcwd(), 'pre') 11 | path_final = os.path.join(os.getcwd(), 'final') 12 | 13 | 14 | def merge(boxes): 15 | """ 16 | 生成的边框可能不止只有一个，需要将边框合并 17 | """ 18 | x, y, h, w = boxes[0] 19 | # x 和 y 应该是整个 boxes 里面最小的值 20 | if len(boxes) > 1: 21 | for tmp in boxes: 22 | x_tmp, y_tmp, h_tmp, w_tmp = tmp 23 | if x > x_tmp: 24 | x_max = x_tmp + w_tmp if x_tmp + w_tmp > x + w else x + w 25 | x = x_tmp 26 | w = x_max - x 27 | if y > y_tmp: 28 | y_max = y_tmp + h_tmp if y_tmp + h_tmp > y + h else y + h 29 | y = y_tmp 30 | h = y_max - y 31 | return tuple((x, y, h, w)) 32 | 33 | 34 | def get_box(png_img): 35 | """ 36 | 获取矩形边框最终返回一个元组 (x,y,h,w)，分别对应矩形左上角的坐标和矩形的高和宽 37 | """ 38 | r, g, b , a = cv2.split(png_img) 39 | gray_img = a 40 | th, binary = cv2.threshold(gray_img, 127 , 255, cv2.THRESH_BINARY) # 二值化 41 | # cv2.imshow("name", binary) 42 | # cv2.waitKey(0) 43 | contours, hierarchy = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # 得到轮廓列表 contours 44 | bounding_boxes = merge([cv2.boundingRect(cnt) for cnt in contours]) # 轮廓合并 45 | # print(bounding_boxes) 46 | return bounding_boxes 47 | 48 | 49 | def get_box_2(png_img): 50 | """ 51 | 不用 opencv 内置算法生成矩形了，改用自己的算法（for 循环） 52 | """ 53 | _, _, _, a = cv2.split(png_img) 54 | _, a = cv2.threshold(a, 127, 255, cv2.THRESH_BINARY) 55 | # 将 r，g，b 通道丢弃，只留下透明度通道 56 | # cv2.imshow("name", a) 57 | # cv2.waitKey(0) 58 | # 在透明度矩阵中，0 代表完全透明 59 | height,width=a.shape # 高和宽 60 | f=0 61 | tmp1 = 0 62 | 63 | """ 64 | 获取上下 65 | """ 66 | for tmp1 in range(0,height): 67 | tmp_a_high= a[tmp1:tmp1+1,:][0] 68 | for tmp2 in range(width): 69 | # a = tmp_a_low[tmp2] 70 | if tmp_a_high[tmp2]!=0: 71 | f=1 72 | if f == 1: 73 | break 74 | delta_y_high = tmp1 + 1 75 | f = 0 76 | for tmp1 in range(height,-1, -1): 77 | tmp_a_low= a[tmp1-1:tmp1+1,:][0] 78 | for tmp2 in range(width): 79 | # a = tmp_a_low[tmp2] 80 | if tmp_a_low[tmp2]!=0: 81 | f=1 82 | if f == 1: 83 | break 84 | delta_y_bottom = height - tmp1 + 3 85 | """ 86 | 获取左右 87 | """ 88 | f = 0 89 | for tmp1 in range(width): 90 | tmp_a_left = a[:, tmp1:tmp1+1] 91 | for tmp2 in range(height): 92 | if tmp_a_left[tmp2] != 0: 93 | f = 1 94 | if f==1: 95 | break 96 | delta_x_left = tmp1 + 1 97 | f = 0 98 | for tmp1 in range(width, -1, -1): 99 | tmp_a_left = a[:, tmp1-1:tmp1] 100 | for tmp2 in range(height): 101 | if tmp_a_left[tmp2] != 0: 102 | f = 1 103 | if f==1: 104 | break 105 | delta_x_right = width - tmp1 + 1 106 | return delta_y_high, delta_y_bottom, delta_x_left, delta_x_right 107 | 108 | 109 | def move(input_image): 110 | """ 111 | 裁剪主函数，输入一张 png 图像，该图像周围是透明的 112 | """ 113 | png_img = input_image # 获取图像 114 | 115 | height, width, channels = png_img.shape # 高 y、宽 x 116 | y_low,y_high, _, _ = get_box_pro(png_img, model=2) # for 循环 117 | base = np.zeros((y_high, width, channels),dtype=np.uint8) # for 循环 118 | png_img = png_img[0:height - y_high, :, :] # for 循环 119 | png_img = np.concatenate((base, png_img), axis=0) 120 | return png_img, y_high 121 | 122 | 123 | def main(): 124 | if not os.path.exists(path_pre): 125 | os.makedirs(path_pre) 126 | if not os.path.exists(path_final): 127 | os.makedirs(path_final) 128 | for name in os.listdir(path_pre): 129 | pass 130 | # move(name) 131 | 132 | 133 | if __name__ == "__main__": 134 | main() 135 | -------------------------------------------------------------------------------- /hivision/creator/photo_adjuster.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding: utf-8 -*- 3 | r""" 4 | @DATE: 2024/9/5 20:02 5 | @File: photo_adjuster.py 6 | @IDE: pycharm 7 | @Description: 8 | 证件照调整 9 | """ 10 | from .context import Context 11 | from .layout_calculator import generate_layout_array 12 | import hivision.creator.utils as U 13 | import numpy as np 14 | import math 15 | import cv2 16 | 17 | 18 | def adjust_photo(ctx: Context): 19 | # Step1. 准备人脸参数 20 | face_rect = ctx.face["rectangle"] 21 | standard_size = ctx.params.size 22 | params = ctx.params 23 | x, y = face_rect[0], face_rect[1] 24 | w, h = face_rect[2], face_rect[3] 25 | height, width = ctx.matting_image.shape[:2] 26 | width_height_ratio = standard_size[0] / standard_size[1] 27 | # Step2. 计算高级参数 28 | face_center = (x + w / 2, y + h / 2) # 面部中心坐标 29 | face_measure = w * h # 面部面积 30 | crop_measure = ( 31 | face_measure / params.head_measure_ratio 32 | ) # 裁剪框面积：为面部面积的 5 倍 33 | resize_ratio = crop_measure / (standard_size[0] * standard_size[1]) # 裁剪框缩放率 34 | resize_ratio_single = math.sqrt( 35 | resize_ratio 36 | ) # 长和宽的缩放率（resize_ratio 的开方） 37 | crop_size = ( 38 | int(standard_size[0] * resize_ratio_single), 39 | int(standard_size[1] * resize_ratio_single), 40 | ) # 裁剪框大小 41 | 42 | # 裁剪框的定位信息 43 | x1 = int(face_center[0] - crop_size[1] / 2) 44 | y1 = int(face_center[1] - crop_size[0] * params.head_height_ratio) 45 | y2 = y1 + crop_size[0] 46 | x2 = x1 + crop_size[1] 47 | 48 | # Step3, 裁剪框的调整 49 | cut_image = IDphotos_cut(x1, y1, x2, y2, ctx.matting_image) 50 | cut_image = cv2.resize(cut_image, (crop_size[1], crop_size[0])) 51 | y_top, y_bottom, x_left, x_right = U.get_box( 52 | cut_image.astype(np.uint8), model=2, correction_factor=0 53 | ) # 得到 cut_image 中人像的上下左右距离信息 54 | 55 | # Step5. 判定 cut_image 中的人像是否处于合理的位置，若不合理，则处理数据以便之后调整位置 56 | # 检测人像与裁剪框左边或右边是否存在空隙 57 | if x_left > 0 or x_right > 0: 58 | status_left_right = 1 59 | cut_value_top = int( 60 | ((x_left + x_right) * width_height_ratio) / 2 61 | ) # 减去左右，为了保持比例，上下也要相应减少 cut_value_top 62 | else: 63 | status_left_right = 0 64 | cut_value_top = 0 65 | 66 | """ 67 | 检测人头顶与照片的顶部是否在合适的距离内： 68 | - status==0: 距离合适，无需移动 69 | - status=1: 距离过大，人像应向上移动 70 | - status=2: 距离过小，人像应向下移动 71 | """ 72 | status_top, move_value = U.detect_distance( 73 | y_top - cut_value_top, 74 | crop_size[0], 75 | max=params.head_top_range[0], 76 | min=params.head_top_range[1], 77 | ) 78 | 79 | # Step6. 对照片的第二轮裁剪 80 | if status_left_right == 0 and status_top == 0: 81 | result_image = cut_image 82 | else: 83 | result_image = IDphotos_cut( 84 | x1 + x_left, 85 | y1 + cut_value_top + status_top * move_value, 86 | x2 - x_right, 87 | y2 - cut_value_top + status_top * move_value, 88 | ctx.matting_image, 89 | ) 90 | 91 | # 换装参数准备 92 | relative_x = x - (x1 + x_left) 93 | relative_y = y - (y1 + cut_value_top + status_top * move_value) 94 | 95 | # Step7. 当照片底部存在空隙时，下拉至底部 96 | result_image, y_high = move(result_image.astype(np.uint8)) 97 | relative_y = relative_y + y_high # 更新换装参数 98 | 99 | # Step8. 标准照与高清照转换 100 | result_image_standard = standard_photo_resize(result_image, standard_size) 101 | result_image_hd, resize_ratio_max = resize_image_by_min( 102 | result_image, esp=max(600, standard_size[1]) 103 | ) 104 | 105 | # Step9. 参数准备 - 为换装服务 106 | clothing_params = { 107 | "relative_x": relative_x * resize_ratio_max, 108 | "relative_y": relative_y * resize_ratio_max, 109 | "w": w * resize_ratio_max, 110 | "h": h * resize_ratio_max, 111 | } 112 | 113 | # Step7. 排版照参数获取 114 | typography_arr, typography_rotate = generate_layout_array( 115 | input_height=standard_size[0], input_width=standard_size[1] 116 | ) 117 | 118 | return ( 119 | result_image_hd, 120 | result_image_standard, 121 | clothing_params, 122 | { 123 | "arr": typography_arr, 124 | "rotate": typography_rotate, 125 | }, 126 | ) 127 | 128 | 129 | def IDphotos_cut(x1, y1, x2, y2, img): 130 | """ 131 | 在图片上进行滑动裁剪，输入输出为 132 | 输入：一张图片 img，和裁剪框信息 (x1,x2,y1,y2) 133 | 输出：裁剪好的图片，然后裁剪框超出了图像范围，那么将用 0 矩阵补位 134 | ------------------------------------ 135 | x:裁剪框左上的横坐标 136 | y:裁剪框左上的纵坐标 137 | x2:裁剪框右下的横坐标 138 | y2:裁剪框右下的纵坐标 139 | crop_size:裁剪框大小 140 | img:裁剪图像（numpy.array） 141 | output_path:裁剪图片的输出路径 142 | ------------------------------------ 143 | """ 144 | 145 | crop_size = (y2 - y1, x2 - x1) 146 | """ 147 | ------------------------------------ 148 | temp_x_1:裁剪框左边超出图像部分 149 | temp_y_1:裁剪框上边超出图像部分 150 | temp_x_2:裁剪框右边超出图像部分 151 | temp_y_2:裁剪框下边超出图像部分 152 | ------------------------------------ 153 | """ 154 | temp_x_1 = 0 155 | temp_y_1 = 0 156 | temp_x_2 = 0 157 | temp_y_2 = 0 158 | 159 | if y1 < 0: 160 | temp_y_1 = abs(y1) 161 | y1 = 0 162 | if y2 > img.shape[0]: 163 | temp_y_2 = y2 164 | y2 = img.shape[0] 165 | temp_y_2 = temp_y_2 - y2 166 | 167 | if x1 < 0: 168 | temp_x_1 = abs(x1) 169 | x1 = 0 170 | if x2 > img.shape[1]: 171 | temp_x_2 = x2 172 | x2 = img.shape[1] 173 | temp_x_2 = temp_x_2 - x2 174 | 175 | # 生成一张全透明背景 176 | background_bgr = np.full((crop_size[0], crop_size[1]), 255, dtype=np.uint8) 177 | background_a = np.full((crop_size[0], crop_size[1]), 0, dtype=np.uint8) 178 | background = cv2.merge( 179 | (background_bgr, background_bgr, background_bgr, background_a) 180 | ) 181 | 182 | background[ 183 | temp_y_1 : crop_size[0] - temp_y_2, temp_x_1 : crop_size[1] - temp_x_2 184 | ] = img[y1:y2, x1:x2] 185 | 186 | return background 187 | 188 | 189 | def move(input_image): 190 | """ 191 | 裁剪主函数，输入一张 png 图像，该图像周围是透明的 192 | """ 193 | png_img = input_image # 获取图像 194 | 195 | height, width, channels = png_img.shape # 高 y、宽 x 196 | y_low, y_high, _, _ = U.get_box(png_img, model=2) # for 循环 197 | base = np.zeros((y_high, width, channels), dtype=np.uint8) # for 循环 198 | png_img = png_img[0 : height - y_high, :, :] # for 循环 199 | png_img = np.concatenate((base, png_img), axis=0) 200 | return png_img, y_high 201 | 202 | 203 | def standard_photo_resize(input_image: np.array, size): 204 | """ 205 | input_image: 输入图像，即高清照 206 | size: 标准照的尺寸 207 | """ 208 | resize_ratio = input_image.shape[0] / size[0] 209 | resize_item = int(round(input_image.shape[0] / size[0])) 210 | if resize_ratio >= 2: 211 | for i in range(resize_item - 1): 212 | if i == 0: 213 | result_image = cv2.resize( 214 | input_image, 215 | (size[1] * (resize_item - i - 1), size[0] * (resize_item - i - 1)), 216 | interpolation=cv2.INTER_AREA, 217 | ) 218 | else: 219 | result_image = cv2.resize( 220 | result_image, 221 | (size[1] * (resize_item - i - 1), size[0] * (resize_item - i - 1)), 222 | interpolation=cv2.INTER_AREA, 223 | ) 224 | else: 225 | result_image = cv2.resize( 226 | input_image, (size[1], size[0]), interpolation=cv2.INTER_AREA 227 | ) 228 | 229 | return result_image 230 | 231 | 232 | def resize_image_by_min(input_image, esp=600): 233 | """ 234 | 将图像缩放为最短边至少为 esp 的图像。 235 | :param input_image: 输入图像（OpenCV 矩阵） 236 | :param esp: 缩放后的最短边长 237 | :return: 缩放后的图像，缩放倍率 238 | """ 239 | height, width = input_image.shape[0], input_image.shape[1] 240 | min_border = min(height, width) 241 | if min_border < esp: 242 | if height >= width: 243 | new_width = esp 244 | new_height = height * esp // width 245 | else: 246 | new_height = esp 247 | new_width = width * esp // height 248 | 249 | return ( 250 | cv2.resize( 251 | input_image, (new_width, new_height), interpolation=cv2.INTER_AREA 252 | ), 253 | new_height / height, 254 | ) 255 | 256 | else: 257 | return input_image, 1 258 | -------------------------------------------------------------------------------- /hivision/creator/retinaface/__init__.py: -------------------------------------------------------------------------------- 1 | from .inference import retinaface_detect_faces 2 | -------------------------------------------------------------------------------- /hivision/creator/retinaface/box_utils.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | 3 | 4 | def decode(loc, priors, variances): 5 | """Decode locations from predictions using priors to undo 6 | the encoding we did for offset regression at train time. 7 | Args: 8 | loc (tensor): location predictions for loc layers, 9 | Shape: [num_priors,4] 10 | priors (tensor): Prior boxes in center-offset form. 11 | Shape: [num_priors,4]. 12 | variances: (list[float]) Variances of priorboxes 13 | Return: 14 | decoded bounding box predictions 15 | """ 16 | 17 | boxes = None 18 | 19 | boxes = np.concatenate( 20 | ( 21 | priors[:, :2] + loc[:, :2] * variances[0] * priors[:, 2:], 22 | priors[:, 2:] * np.exp(loc[:, 2:] * variances[1]), 23 | ), 24 | axis=1, 25 | ) 26 | 27 | boxes[:, :2] -= boxes[:, 2:] / 2 28 | boxes[:, 2:] += boxes[:, :2] 29 | return boxes 30 | 31 | 32 | def decode_landm(pre, priors, variances): 33 | """Decode landm from predictions using priors to undo 34 | the encoding we did for offset regression at train time. 35 | Args: 36 | pre (tensor): landm predictions for loc layers, 37 | Shape: [num_priors,10] 38 | priors (tensor): Prior boxes in center-offset form. 39 | Shape: [num_priors,4]. 40 | variances: (list[float]) Variances of priorboxes 41 | Return: 42 | decoded landm predictions 43 | """ 44 | landms = None 45 | 46 | landms = np.concatenate( 47 | ( 48 | priors[:, :2] + pre[:, :2] * variances[0] * priors[:, 2:], 49 | priors[:, :2] + pre[:, 2:4] * variances[0] * priors[:, 2:], 50 | priors[:, :2] + pre[:, 4:6] * variances[0] * priors[:, 2:], 51 | priors[:, :2] + pre[:, 6:8] * variances[0] * priors[:, 2:], 52 | priors[:, :2] + pre[:, 8:10] * variances[0] * priors[:, 2:], 53 | ), 54 | axis=1, 55 | ) 56 | 57 | return landms 58 | -------------------------------------------------------------------------------- /hivision/creator/retinaface/inference.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import cv2 3 | import onnxruntime 4 | from hivision.creator.retinaface.box_utils import decode, decode_landm 5 | from hivision.creator.retinaface.prior_box import PriorBox 6 | 7 | 8 | def py_cpu_nms(dets, thresh): 9 | """Pure Python NMS baseline.""" 10 | x1 = dets[:, 0] 11 | y1 = dets[:, 1] 12 | x2 = dets[:, 2] 13 | y2 = dets[:, 3] 14 | scores = dets[:, 4] 15 | 16 | areas = (x2 - x1 + 1) * (y2 - y1 + 1) 17 | order = scores.argsort()[::-1] 18 | 19 | keep = [] 20 | while order.size > 0: 21 | i = order[0] 22 | keep.append(i) 23 | xx1 = np.maximum(x1[i], x1[order[1:]]) 24 | yy1 = np.maximum(y1[i], y1[order[1:]]) 25 | xx2 = np.minimum(x2[i], x2[order[1:]]) 26 | yy2 = np.minimum(y2[i], y2[order[1:]]) 27 | 28 | w = np.maximum(0.0, xx2 - xx1 + 1) 29 | h = np.maximum(0.0, yy2 - yy1 + 1) 30 | inter = w * h 31 | ovr = inter / (areas[i] + areas[order[1:]] - inter) 32 | 33 | inds = np.where(ovr <= thresh)[0] 34 | order = order[inds + 1] 35 | 36 | return keep 37 | 38 | 39 | # 替换掉 argparse 的部分，直接使用普通变量 40 | network = "resnet50" 41 | use_cpu = False 42 | confidence_threshold = 0.8 43 | top_k = 5000 44 | nms_threshold = 0.2 45 | keep_top_k = 750 46 | save_image = True 47 | vis_thres = 0.6 48 | 49 | ONNX_DEVICE = ( 50 | "CUDAExecutionProvider" 51 | if onnxruntime.get_device() == "GPU" 52 | else "CPUExecutionProvider" 53 | ) 54 | 55 | 56 | def load_onnx_model(checkpoint_path, set_cpu=False): 57 | providers = ( 58 | ["CUDAExecutionProvider", "CPUExecutionProvider"] 59 | if ONNX_DEVICE == "CUDAExecutionProvider" 60 | else ["CPUExecutionProvider"] 61 | ) 62 | 63 | if set_cpu: 64 | sess = onnxruntime.InferenceSession( 65 | checkpoint_path, providers=["CPUExecutionProvider"] 66 | ) 67 | else: 68 | try: 69 | sess = onnxruntime.InferenceSession(checkpoint_path, providers=providers) 70 | except Exception as e: 71 | if ONNX_DEVICE == "CUDAExecutionProvider": 72 | print(f"Failed to load model with CUDAExecutionProvider: {e}") 73 | print("Falling back to CPUExecutionProvider") 74 | # 尝试使用CPU加载模型 75 | sess = onnxruntime.InferenceSession( 76 | checkpoint_path, providers=["CPUExecutionProvider"] 77 | ) 78 | else: 79 | raise e # 如果是CPU执行失败，重新抛出异常 80 | 81 | return sess 82 | 83 | 84 | def retinaface_detect_faces(image, model_path: str, sess=None): 85 | cfg = { 86 | "name": "Resnet50", 87 | "min_sizes": [[16, 32], [64, 128], [256, 512]], 88 | "steps": [8, 16, 32], 89 | "variance": [0.1, 0.2], 90 | "clip": False, 91 | "loc_weight": 2.0, 92 | "gpu_train": True, 93 | "batch_size": 24, 94 | "ngpu": 4, 95 | "epoch": 100, 96 | "decay1": 70, 97 | "decay2": 90, 98 | "image_size": 840, 99 | "pretrain": True, 100 | "return_layers": {"layer2": 1, "layer3": 2, "layer4": 3}, 101 | "in_channel": 256, 102 | "out_channel": 256, 103 | } 104 | 105 | # Load ONNX model 106 | if sess is None: 107 | retinaface = load_onnx_model(model_path, set_cpu=False) 108 | else: 109 | retinaface = sess 110 | 111 | resize = 1 112 | 113 | # Read and preprocess the image 114 | img_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) 115 | img = np.float32(img_rgb) 116 | 117 | im_height, im_width, _ = img.shape 118 | scale = np.array([img.shape[1], img.shape[0], img.shape[1], img.shape[0]]) 119 | img -= (104, 117, 123) 120 | img = img.transpose(2, 0, 1) 121 | img = np.expand_dims(img, axis=0) 122 | 123 | # Run the model 124 | inputs = {"input": img} 125 | loc, conf, landms = retinaface.run(None, inputs) 126 | 127 | priorbox = PriorBox(cfg, image_size=(im_height, im_width)) 128 | priors = priorbox.forward() 129 | 130 | prior_data = priors 131 | 132 | boxes = decode(np.squeeze(loc, axis=0), prior_data, cfg["variance"]) 133 | boxes = boxes * scale / resize 134 | scores = np.squeeze(conf, axis=0)[:, 1] 135 | 136 | landms = decode_landm(np.squeeze(landms.data, axis=0), prior_data, cfg["variance"]) 137 | 138 | scale1 = np.array( 139 | [ 140 | img.shape[3], 141 | img.shape[2], 142 | img.shape[3], 143 | img.shape[2], 144 | img.shape[3], 145 | img.shape[2], 146 | img.shape[3], 147 | img.shape[2], 148 | img.shape[3], 149 | img.shape[2], 150 | ] 151 | ) 152 | landms = landms * scale1 / resize 153 | 154 | # ignore low scores 155 | inds = np.where(scores > confidence_threshold)[0] 156 | boxes = boxes[inds] 157 | landms = landms[inds] 158 | scores = scores[inds] 159 | 160 | # keep top-K before NMS 161 | order = scores.argsort()[::-1][:top_k] 162 | boxes = boxes[order] 163 | landms = landms[order] 164 | scores = scores[order] 165 | 166 | # do NMS 167 | dets = np.hstack((boxes, scores[:, np.newaxis])).astype(np.float32, copy=False) 168 | keep = py_cpu_nms(dets, nms_threshold) 169 | dets = dets[keep, :] 170 | landms = landms[keep] 171 | 172 | # keep top-K faster NMS 173 | dets = dets[:keep_top_k, :] 174 | landms = landms[:keep_top_k, :] 175 | 176 | dets = np.concatenate((dets, landms), axis=1) 177 | 178 | return dets, retinaface 179 | 180 | 181 | if __name__ == "__main__": 182 | import gradio as gr 183 | 184 | # Create Gradio interface 185 | iface = gr.Interface( 186 | fn=retinaface_detect_faces, 187 | inputs=[ 188 | gr.Image( 189 | type="numpy", label="上传图片", height=400 190 | ), # Set the height to 400 191 | gr.Textbox(value="./FaceDetector.onnx", label="ONNX模型路径"), 192 | ], 193 | outputs=gr.Number(label="检测到的人脸数量"), 194 | title="人脸检测", 195 | description="上传图片并提供ONNX模型路径以检测人脸数量。", 196 | ) 197 | 198 | # Launch the Gradio app 199 | iface.launch() 200 | -------------------------------------------------------------------------------- /hivision/creator/retinaface/prior_box.py: -------------------------------------------------------------------------------- 1 | from itertools import product as product 2 | import numpy as np 3 | from math import ceil 4 | 5 | 6 | class PriorBox(object): 7 | def __init__(self, cfg, image_size=None): 8 | super(PriorBox, self).__init__() 9 | self.min_sizes = cfg["min_sizes"] 10 | self.steps = cfg["steps"] 11 | self.clip = cfg["clip"] 12 | self.image_size = image_size 13 | self.feature_maps = [ 14 | [ceil(self.image_size[0] / step), ceil(self.image_size[1] / step)] 15 | for step in self.steps 16 | ] 17 | self.name = "s" 18 | 19 | def forward(self): 20 | anchors = [] 21 | for k, f in enumerate(self.feature_maps): 22 | min_sizes = self.min_sizes[k] 23 | for i, j in product(range(f[0]), range(f[1])): 24 | for min_size in min_sizes: 25 | s_kx = min_size / self.image_size[1] 26 | s_ky = min_size / self.image_size[0] 27 | dense_cx = [ 28 | x * self.steps[k] / self.image_size[1] for x in [j + 0.5] 29 | ] 30 | dense_cy = [ 31 | y * self.steps[k] / self.image_size[0] for y in [i + 0.5] 32 | ] 33 | for cy, cx in product(dense_cy, dense_cx): 34 | anchors += [cx, cy, s_kx, s_ky] 35 | 36 | output = np.array(anchors).reshape(-1, 4) 37 | 38 | if self.clip: 39 | output = np.clip(output, 0, 1) 40 | 41 | return output 42 | -------------------------------------------------------------------------------- /hivision/creator/retinaface/weights/.gitkeep: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/hivision/creator/retinaface/weights/.gitkeep -------------------------------------------------------------------------------- /hivision/creator/rotation_adjust.py: -------------------------------------------------------------------------------- 1 | """ 2 | 人脸旋转矫正模块 3 | 4 | 本模块提供了用于旋转图像的函数，主要用于人脸旋转矫正。 5 | 包含了处理3通道和4通道图像的旋转函数。 6 | """ 7 | 8 | import cv2 9 | import numpy as np 10 | 11 | 12 | def rotate_bound(image: np.ndarray, angle: float, center=None): 13 | """ 14 | 旋转图像而不损失信息的函数 15 | 16 | Args: 17 | image (np.ndarray): 输入图像，3通道numpy数组 18 | angle (float): 旋转角度（度） 19 | center (tuple, optional): 旋转中心坐标，默认为图像中心 20 | 21 | Returns: 22 | tuple: 包含以下元素的元组： 23 | - rotated (np.ndarray): 旋转后的图像 24 | - cos (float): 旋转角度的余弦值 25 | - sin (float): 旋转角度的正弦值 26 | - dW (int): 宽度变化量 27 | - dH (int): 高度变化量 28 | """ 29 | (h, w) = image.shape[:2] 30 | if center is None: 31 | (cX, cY) = (w / 2, h / 2) 32 | else: 33 | (cX, cY) = center 34 | 35 | M = cv2.getRotationMatrix2D((cX, cY), -angle, 1.0) 36 | cos = np.abs(M[0, 0]) 37 | sin = np.abs(M[0, 1]) 38 | 39 | nW = int((h * sin) + (w * cos)) 40 | nH = int((h * cos) + (w * sin)) 41 | 42 | M[0, 2] += (nW / 2) - cX 43 | M[1, 2] += (nH / 2) - cY 44 | 45 | rotated = cv2.warpAffine(image, M, (nW, nH)) 46 | 47 | # 计算偏移量 48 | dW = nW - w 49 | dH = nH - h 50 | 51 | return rotated, cos, sin, dW, dH 52 | 53 | 54 | def rotate_bound_4channels(image: np.ndarray, a: np.ndarray, angle: float, center=None): 55 | """ 56 | 旋转4通道图像的函数 57 | 58 | 这是rotate_bound函数的4通道版本，可以同时处理RGB图像和其对应的alpha通道。 59 | 60 | Args: 61 | image (np.ndarray): 输入的3通道RGB图像 62 | a (np.ndarray): 输入图像的alpha通道 63 | angle (float): 旋转角度（度） 64 | center (tuple, optional): 旋转中心坐标，默认为图像中心 65 | 66 | Returns: 67 | tuple: 包含以下元素的元组： 68 | - input_image (np.ndarray): 旋转后的3通道RGB图像 69 | - result_image (np.ndarray): 旋转后的4通道RGBA图像 70 | - cos (float): 旋转角度的余弦值 71 | - sin (float): 旋转角度的正弦值 72 | - dW (int): 宽度变化量 73 | - dH (int): 高度变化量 74 | """ 75 | input_image, cos, sin, dW, dH = rotate_bound(image, angle, center) 76 | new_a, _, _, _, _ = rotate_bound(a, angle, center) # 对alpha通道进行旋转 77 | b, g, r = cv2.split(input_image) 78 | result_image = cv2.merge((b, g, r, new_a)) # 合并旋转后的RGB通道和alpha通道 79 | 80 | return input_image, result_image, cos, sin, dW, dH 81 | -------------------------------------------------------------------------------- /hivision/creator/tensor2numpy.py: -------------------------------------------------------------------------------- 1 | """ 2 | 作者：林泽毅 3 | 建这个开源库的起源呢，是因为在做 onnx 推理的时候，需要将原来的 tensor 转换成 numpy.array 4 | 问题是 Tensor 和 Numpy 的矩阵排布逻辑不同 5 | 包括 Tensor 推理经常会进行 Transform，比如 ToTensor,Normalize 等 6 | 就想做一些等价转换的函数。 7 | """ 8 | import numpy as np 9 | 10 | 11 | def NTo_Tensor(array): 12 | """ 13 | :param array: opencv/PIL读取的numpy矩阵 14 | :return:返回一个形如 Tensor 的 numpy 矩阵 15 | Example: 16 | Inputs:array.shape = (512,512,3) 17 | Outputs:output.shape = (3,512,512) 18 | """ 19 | output = array.transpose((2, 0, 1)) 20 | return output 21 | 22 | 23 | def NNormalize(array, mean=np.array([0.5, 0.5, 0.5]), std=np.array([0.5, 0.5, 0.5]), dtype=np.float32): 24 | """ 25 | :param array: opencv/PIL读取的numpy矩阵 26 | mean: 归一化均值，np.array 格式 27 | std: 归一化标准差，np.array 格式 28 | dtype：输出的 numpy 数据格式，一般 onnx 需要 float32 29 | :return:numpy 矩阵 30 | Example: 31 | Inputs:array 为 opencv/PIL 读取的一张图片 32 | mean=np.array([0.5,0.5,0.5]) 33 | std=np.array([0.5,0.5,0.5]) 34 | dtype=np.float32 35 | Outputs:output 为归一化后的 numpy 矩阵 36 | """ 37 | im = array / 255.0 38 | im = np.divide(np.subtract(im, mean), std) 39 | output = np.asarray(im, dtype=dtype) 40 | 41 | return output 42 | 43 | 44 | def NUnsqueeze(array, axis=0): 45 | """ 46 | :param array: opencv/PIL读取的numpy矩阵 47 | axis：要增加的维度 48 | :return:numpy 矩阵 49 | Example: 50 | Inputs:array 为 opencv/PIL 读取的一张图片，array.shape 为 [512,512,3] 51 | axis=0 52 | Outputs:output 为 array 在第 0 维增加一个维度，shape 转为 [1,512,512,3] 53 | """ 54 | if axis == 0: 55 | output = array[None, :, :, :] 56 | elif axis == 1: 57 | output = array[:, None, :, :] 58 | elif axis == 2: 59 | output = array[:, :, None, :] 60 | else: 61 | output = array[:, :, :, None] 62 | 63 | return output 64 | -------------------------------------------------------------------------------- /hivision/creator/utils.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding: utf-8 -*- 3 | r""" 4 | @DATE: 2024/9/5 19:25 5 | @File: utils.py 6 | @IDE: pycharm 7 | @Description: 8 | 通用图像处理工具 9 | """ 10 | import cv2 11 | import numpy as np 12 | 13 | 14 | def resize_image_esp(input_image, esp=2000): 15 | """ 16 | 输入： 17 | input_path：numpy 图片 18 | esp：限制的最大边长 19 | """ 20 | # resize 函数=>可以让原图压缩到最大边为 esp 的尺寸 (不改变比例) 21 | width = input_image.shape[0] 22 | 23 | length = input_image.shape[1] 24 | max_num = max(width, length) 25 | 26 | if max_num > esp: 27 | print("Image resizing...") 28 | if width == max_num: 29 | length = int((esp / width) * length) 30 | width = esp 31 | 32 | else: 33 | width = int((esp / length) * width) 34 | length = esp 35 | print(length, width) 36 | im_resize = cv2.resize( 37 | input_image, (length, width), interpolation=cv2.INTER_AREA 38 | ) 39 | return im_resize 40 | else: 41 | return input_image 42 | 43 | 44 | def get_box( 45 | image: np.ndarray, 46 | model: int = 1, 47 | correction_factor=None, 48 | thresh: int = 127, 49 | ): 50 | """ 51 | 本函数能够实现输入一张四通道图像，返回图像中最大连续非透明面积的区域的矩形坐标 52 | 本函数将采用 opencv 内置函数来解析整个图像的 mask，并提供一些参数，用于读取图像的位置信息 53 | Args: 54 | image: 四通道矩阵图像 55 | model: 返回值模式 56 | correction_factor: 提供一些边缘扩张接口，输入格式为 list 或者 int:[up, down, left, right]。 57 | 举个例子，假设我们希望剪切出的矩形框左边能够偏左 1 个像素，则输入 [0, 0, 1, 0]； 58 | 如果希望右边偏右 1 个像素，则输入 [0, 0, 0, 1] 59 | 如果输入为 int，则默认只会对左右两边做拓展，比如输入 2，则和 [0, 0, 2, 2] 是等效的 60 | thresh: 二值化阈值，为了保持一些羽化效果，thresh 必须要小 61 | Returns: 62 | model 为 1 时，将会返回切割出的矩形框的四个坐标点信息 63 | model 为 2 时，将会返回矩形框四边相距于原图四边的距离 64 | """ 65 | # ------------ 数据格式规范部分 -------------- # 66 | # 输入必须为四通道 67 | if correction_factor is None: 68 | correction_factor = [0, 0, 0, 0] 69 | if not isinstance(image, np.ndarray) or len(cv2.split(image)) != 4: 70 | raise TypeError("输入的图像必须为四通道 np.ndarray 类型矩阵！") 71 | # correction_factor 规范化 72 | if isinstance(correction_factor, int): 73 | correction_factor = [0, 0, correction_factor, correction_factor] 74 | elif not isinstance(correction_factor, list): 75 | raise TypeError("correction_factor 必须为 int 或者 list 类型！") 76 | # ------------ 数据格式规范完毕 -------------- # 77 | # 分离 mask 78 | _, _, _, mask = cv2.split(image) 79 | # mask 二值化处理 80 | _, mask = cv2.threshold(mask, thresh=thresh, maxval=255, type=0) 81 | contours, hierarchy = cv2.findContours(mask, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) 82 | temp = np.ones(image.shape, np.uint8) * 255 83 | cv2.drawContours(temp, contours, -1, (0, 0, 255), -1) 84 | contours_area = [] 85 | for cnt in contours: 86 | contours_area.append(cv2.contourArea(cnt)) 87 | idx = contours_area.index(max(contours_area)) 88 | x, y, w, h = cv2.boundingRect(contours[idx]) # 框出图像 89 | # ------------ 开始输出数据 -------------- # 90 | height, width, _ = image.shape 91 | y_up = y - correction_factor[0] if y - correction_factor[0] >= 0 else 0 92 | y_down = ( 93 | y + h + correction_factor[1] 94 | if y + h + correction_factor[1] < height 95 | else height - 1 96 | ) 97 | x_left = x - correction_factor[2] if x - correction_factor[2] >= 0 else 0 98 | x_right = ( 99 | x + w + correction_factor[3] 100 | if x + w + correction_factor[3] < width 101 | else width - 1 102 | ) 103 | if model == 1: 104 | # model=1，将会返回切割出的矩形框的四个坐标点信息 105 | return [y_up, y_down, x_left, x_right] 106 | elif model == 2: 107 | # model=2, 将会返回矩形框四边相距于原图四边的距离 108 | return [y_up, height - y_down, x_left, width - x_right] 109 | else: 110 | raise EOFError("请选择正确的模式！") 111 | 112 | 113 | def detect_distance(value, crop_height, max=0.06, min=0.04): 114 | """ 115 | 检测人头顶与照片顶部的距离是否在适当范围内。 116 | 输入：与顶部的差值 117 | 输出：(status, move_value) 118 | status=0 不动 119 | status=1 人脸应向上移动（裁剪框向下移动） 120 | status-2 人脸应向下移动（裁剪框向上移动） 121 | --------------------------------------- 122 | value：头顶与照片顶部的距离 123 | crop_height: 裁剪框的高度 124 | max: 距离的最大值 125 | min: 距离的最小值 126 | --------------------------------------- 127 | """ 128 | value = value / crop_height # 头顶往上的像素占图像的比例 129 | if min <= value <= max: 130 | return 0, 0 131 | elif value > max: 132 | # 头顶往上的像素比例高于 max 133 | move_value = value - max 134 | move_value = int(move_value * crop_height) 135 | # print("上移{}".format(move_value)) 136 | return 1, move_value 137 | else: 138 | # 头顶往上的像素比例低于 min 139 | move_value = min - value 140 | move_value = int(move_value * crop_height) 141 | # print("下移{}".format(move_value)) 142 | return -1, move_value 143 | 144 | 145 | def cutting_rect_pan( 146 | x1, y1, x2, y2, width, height, L1, L2, L3, clockwise, standard_size 147 | ): 148 | """ 149 | 本函数的功能是对旋转矫正结果图的裁剪框进行修正 ———— 解决"旋转三角形"现象。 150 | Args: 151 | - x1: int, 裁剪框左上角的横坐标 152 | - y1: int, 裁剪框左上角的纵坐标 153 | - x2: int, 裁剪框右下角的横坐标 154 | - y2: int, 裁剪框右下角的纵坐标 155 | - width: int, 待裁剪图的宽度 156 | - height:int, 待裁剪图的高度 157 | - L1: CLassObject, 根据旋转点连线所构造函数 158 | - L2: CLassObject, 根据旋转点连线所构造函数 159 | - L3: ClassObject, 一个特殊裁切点的坐标 160 | - clockwise: int, 旋转时针状态 161 | - standard_size: tuple, 标准照的尺寸 162 | 163 | Returns: 164 | - x1: int, 新的裁剪框左上角的横坐标 165 | - y1: int, 新的裁剪框左上角的纵坐标 166 | - x2: int, 新的裁剪框右下角的横坐标 167 | - y2: int, 新的裁剪框右下角的纵坐标 168 | - x_bias: int, 裁剪框横坐标方向上的计算偏置量 169 | - y_bias: int, 裁剪框纵坐标方向上的计算偏置量 170 | """ 171 | # 用于计算的裁剪框坐标x1_cal,x2_cal,y1_cal,y2_cal(如果裁剪框超出了图像范围，则缩小直至在范围内) 172 | x1_std = x1 if x1 > 0 else 0 173 | x2_std = x2 if x2 < width else width 174 | # y1_std = y1 if y1 > 0 else 0 175 | y2_std = y2 if y2 < height else height 176 | 177 | # 初始化x和y的计算偏置项x_bias和y_bias 178 | x_bias = 0 179 | y_bias = 0 180 | 181 | # 如果顺时针偏转 182 | if clockwise == 1: 183 | if y2 > L1.forward_x(x1_std): 184 | y_bias = int(-(y2_std - L1.forward_x(x1_std))) 185 | if y2 > L2.forward_x(x2_std): 186 | x_bias = int(-(x2_std - L2.forward_y(y2_std))) 187 | x2 = x2_std + x_bias 188 | if x1 < L3.x: 189 | x1 = L3.x 190 | # 如果逆时针偏转 191 | else: 192 | if y2 > L1.forward_x(x1_std): 193 | x_bias = int(L1.forward_y(y2_std) - x1_std) 194 | if y2 > L2.forward_x(x2_std): 195 | y_bias = int(-(y2_std - L2.forward_x(x2_std))) 196 | x1 = x1_std + x_bias 197 | if x2 > L3.x: 198 | x2 = L3.x 199 | 200 | # 计算裁剪框的y的变化 201 | y2 = int(y2_std + y_bias) 202 | new_cut_width = x2 - x1 203 | new_cut_height = int(new_cut_width / standard_size[1] * standard_size[0]) 204 | y1 = y2 - new_cut_height 205 | 206 | return x1, y1, x2, y2, x_bias, y_bias 207 | -------------------------------------------------------------------------------- /hivision/creator/weights/.gitkeep: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/hivision/creator/weights/.gitkeep -------------------------------------------------------------------------------- /hivision/error.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding: utf-8 -*- 3 | r""" 4 | @DATE: 2024/9/5 18:32 5 | @File: error.py 6 | @IDE: pycharm 7 | @Description: 8 | 错误处理 9 | """ 10 | 11 | 12 | class FaceError(Exception): 13 | def __init__(self, err, face_num): 14 | """ 15 | 证件照人脸错误，此时人脸检测失败，可能是没有检测到人脸或者检测到多个人脸 16 | Args: 17 | err: 错误描述 18 | face_num: 告诉此时识别到的人像个数 19 | """ 20 | super().__init__(err) 21 | self.face_num = face_num 22 | 23 | 24 | class APIError(Exception): 25 | def __init__(self, err, status_code): 26 | """ 27 | API错误 28 | Args: 29 | err: 错误描述 30 | status_code: 告诉此时的错误状态码 31 | """ 32 | super().__init__(err) 33 | self.status_code = status_code 34 | -------------------------------------------------------------------------------- /hivision/plugin/beauty/__init__.py: -------------------------------------------------------------------------------- 1 | from .beauty_tools import BeautyTools 2 | -------------------------------------------------------------------------------- /hivision/plugin/beauty/base_adjust.py: -------------------------------------------------------------------------------- 1 | """ 2 | 亮度、对比度、锐化、饱和度调整模块 3 | """ 4 | 5 | import cv2 6 | import numpy as np 7 | 8 | 9 | def adjust_brightness_contrast_sharpen_saturation( 10 | image, 11 | brightness_factor=0, 12 | contrast_factor=0, 13 | sharpen_strength=0, 14 | saturation_factor=0, 15 | ): 16 | """ 17 | 调整图像的亮度、对比度、锐度和饱和度。 18 | 19 | 参数: 20 | image (numpy.ndarray): 输入的图像数组。 21 | brightness_factor (float): 亮度调整因子。大于0增加亮度，小于0降低亮度。 22 | contrast_factor (float): 对比度调整因子。大于0增加对比度，小于0降低对比度。 23 | sharpen_strength (float): 锐化强度。 24 | saturation_factor (float): 饱和度调整因子。大于0增加饱和度，小于0降低饱和度。 25 | 26 | 返回: 27 | numpy.ndarray: 调整后的图像。 28 | """ 29 | if ( 30 | brightness_factor == 0 31 | and contrast_factor == 0 32 | and sharpen_strength == 0 33 | and saturation_factor == 0 34 | ): 35 | return image.copy() 36 | 37 | adjusted_image = image.copy() 38 | 39 | # 调整饱和度 40 | if saturation_factor != 0: 41 | adjusted_image = adjust_saturation(adjusted_image, saturation_factor) 42 | 43 | # 调整亮度和对比度 44 | alpha = 1.0 + (contrast_factor / 100.0) 45 | beta = brightness_factor 46 | adjusted_image = cv2.convertScaleAbs(adjusted_image, alpha=alpha, beta=beta) 47 | 48 | # 增强锐化 49 | adjusted_image = sharpen_image(adjusted_image, sharpen_strength) 50 | 51 | return adjusted_image 52 | 53 | 54 | def adjust_saturation(image, saturation_factor): 55 | """ 56 | 调整图像的饱和度。 57 | 58 | 参数: 59 | image (numpy.ndarray): 输入的图像数组。 60 | saturation_factor (float): 饱和度调整因子。大于0增加饱和度，小于0降低饱和度。 61 | 62 | 返回: 63 | numpy.ndarray: 调整后的图像。 64 | """ 65 | hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV) 66 | h, s, v = cv2.split(hsv) 67 | s = s.astype(np.float32) 68 | s = s + s * (saturation_factor / 100.0) 69 | s = np.clip(s, 0, 255).astype(np.uint8) 70 | hsv = cv2.merge([h, s, v]) 71 | return cv2.cvtColor(hsv, cv2.COLOR_HSV2BGR) 72 | 73 | 74 | def sharpen_image(image, strength=0): 75 | """ 76 | 对图像进行锐化处理。 77 | 78 | 参数: 79 | image (numpy.ndarray): 输入的图像数组。 80 | strength (float): 锐化强度，范围建议为0-5。0表示不进行锐化。 81 | 82 | 返回: 83 | numpy.ndarray: 锐化后的图像。 84 | """ 85 | print(f"Sharpen strength: {strength}") 86 | if strength == 0: 87 | return image.copy() 88 | 89 | strength = strength * 20 90 | kernel_strength = 1 + (strength / 500) 91 | 92 | kernel = ( 93 | np.array([[-0.5, -0.5, -0.5], [-0.5, 5, -0.5], [-0.5, -0.5, -0.5]]) 94 | * kernel_strength 95 | ) 96 | 97 | sharpened = cv2.filter2D(image, -1, kernel) 98 | sharpened = np.clip(sharpened, 0, 255).astype(np.uint8) 99 | 100 | alpha = strength / 200 101 | blended = cv2.addWeighted(image, 1 - alpha, sharpened, alpha, 0) 102 | 103 | return blended 104 | 105 | 106 | # Gradio接口 107 | def base_adjustment(image, brightness, contrast, sharpen, saturation): 108 | adjusted = adjust_brightness_contrast_sharpen_saturation( 109 | image, brightness, contrast, sharpen, saturation 110 | ) 111 | return adjusted 112 | 113 | 114 | if __name__ == "__main__": 115 | import gradio as gr 116 | 117 | iface = gr.Interface( 118 | fn=base_adjustment, 119 | inputs=[ 120 | gr.Image(label="Input Image", height=400), 121 | gr.Slider( 122 | minimum=-20, 123 | maximum=20, 124 | value=0, 125 | step=1, 126 | label="Brightness", 127 | ), 128 | gr.Slider( 129 | minimum=-100, 130 | maximum=100, 131 | value=0, 132 | step=1, 133 | label="Contrast", 134 | ), 135 | gr.Slider( 136 | minimum=0, 137 | maximum=5, 138 | value=0, 139 | step=1, 140 | label="Sharpen", 141 | ), 142 | gr.Slider( 143 | minimum=-100, 144 | maximum=100, 145 | value=0, 146 | step=1, 147 | label="Saturation", 148 | ), 149 | ], 150 | outputs=gr.Image(label="Adjusted Image"), 151 | title="Image Adjustment", 152 | description="Adjust the brightness, contrast, sharpness, and saturation of an image using sliders.", 153 | ) 154 | iface.launch() 155 | -------------------------------------------------------------------------------- /hivision/plugin/beauty/beauty_tools.py: -------------------------------------------------------------------------------- 1 | """ 2 | @author: cuny 3 | @file: MakeBeautiful.py 4 | @time: 2022/7/7 20:23 5 | @description: 6 | 美颜工具集合文件，作为暴露在外的插件接口 7 | """ 8 | 9 | from .grind_skin import grindSkin 10 | from .whitening import MakeWhiter 11 | from .thin_face import thinFace 12 | import numpy as np 13 | 14 | 15 | def BeautyTools( 16 | input_image: np.ndarray, 17 | landmark, 18 | thinStrength: int, 19 | thinPlace: int, 20 | grindStrength: int, 21 | whiterStrength: int, 22 | ) -> np.ndarray: 23 | """ 24 | 美颜工具的接口函数，用于实现美颜效果 25 | Args: 26 | input_image: 输入的图像 27 | landmark: 瘦脸需要的人脸关键点信息，为fd68返回的第二个参数 28 | thinStrength: 瘦脸强度，为0-10（如果更高其实也没什么问题），当强度为0或者更低时，则不瘦脸 29 | thinPlace: 选择瘦脸区域，为0-2之间的值，越大瘦脸的点越靠下 30 | grindStrength: 磨皮强度，为0-10（如果更高其实也没什么问题），当强度为0或者更低时，则不磨皮 31 | whiterStrength: 美白强度，为0-10（如果更高其实也没什么问题），当强度为0或者更低时，则不美白 32 | Returns: 33 | output_image 输出图像 34 | """ 35 | try: 36 | _, _, _ = input_image.shape 37 | except ValueError: 38 | raise TypeError("输入图像必须为3通道或者4通道!") 39 | # 三通道或者四通道图像 40 | # 首先进行瘦脸 41 | input_image = thinFace( 42 | input_image, landmark, place=thinPlace, strength=thinStrength 43 | ) 44 | # 其次进行磨皮 45 | input_image = grindSkin(src=input_image, strength=grindStrength) 46 | # 最后进行美白 47 | makeWhiter = MakeWhiter() 48 | input_image = makeWhiter.run(input_image, strength=whiterStrength) 49 | return input_image 50 | -------------------------------------------------------------------------------- /hivision/plugin/beauty/grind_skin.py: -------------------------------------------------------------------------------- 1 | # Required Libraries 2 | import cv2 3 | import numpy as np 4 | import gradio as gr 5 | 6 | 7 | def annotate_image(image, grind_degree, detail_degree, strength): 8 | """Annotates the image with parameters in the lower-left corner.""" 9 | font = cv2.FONT_HERSHEY_SIMPLEX 10 | font_scale = 0.5 11 | color = (0, 0, 255) 12 | thickness = 1 13 | line_type = cv2.LINE_AA 14 | 15 | # Text positions 16 | y_offset = 20 17 | x_offset = 10 18 | y_base = image.shape[0] - 10 19 | 20 | # Define each line of the annotation 21 | lines = [ 22 | f"Grind Degree: {grind_degree}", 23 | f"Detail Degree: {detail_degree}", 24 | f"Strength: {strength}", 25 | ] 26 | 27 | # Draw the text lines on the image 28 | for i, line in enumerate(lines): 29 | y_position = y_base - (i * y_offset) 30 | cv2.putText( 31 | image, 32 | line, 33 | (x_offset, y_position), 34 | font, 35 | font_scale, 36 | color, 37 | thickness, 38 | line_type, 39 | ) 40 | 41 | return image 42 | 43 | 44 | def grindSkin(src, grindDegree: int = 3, detailDegree: int = 1, strength: int = 9): 45 | """ 46 | Dest =(Src * (100 - Opacity) + (Src + 2 * GaussBlur(EPFFilter(Src) - Src)) * Opacity) / 100 47 | 人像磨皮方案 48 | Args: 49 | src: 原图 50 | grindDegree: 磨皮程度调节参数 51 | detailDegree: 细节程度调节参数 52 | strength: 融合程度，作为磨皮强度（0 - 10） 53 | 54 | Returns: 55 | 磨皮后的图像 56 | """ 57 | if strength <= 0: 58 | return src 59 | dst = src.copy() 60 | opacity = min(10.0, strength) / 10.0 61 | dx = grindDegree * 5 62 | fc = grindDegree * 12.5 63 | temp1 = cv2.bilateralFilter(src[:, :, :3], dx, fc, fc) 64 | temp2 = cv2.subtract(temp1, src[:, :, :3]) 65 | temp3 = cv2.GaussianBlur(temp2, (2 * detailDegree - 1, 2 * detailDegree - 1), 0) 66 | temp4 = cv2.add(cv2.add(temp3, temp3), src[:, :, :3]) 67 | dst[:, :, :3] = cv2.addWeighted(temp4, opacity, src[:, :, :3], 1 - opacity, 0.0) 68 | return dst 69 | 70 | 71 | def process_image(input_img, grind_degree, detail_degree, strength): 72 | # Reading the image using OpenCV 73 | img = cv2.cvtColor(input_img, cv2.COLOR_RGB2BGR) 74 | # Processing the image 75 | output_img = grindSkin(img, grind_degree, detail_degree, strength) 76 | # Annotating the processed image with parameters 77 | output_img_annotated = annotate_image( 78 | output_img.copy(), grind_degree, detail_degree, strength 79 | ) 80 | # Horizontal stacking of input and processed images 81 | combined_img = cv2.hconcat([img, output_img_annotated]) 82 | # Convert the combined image back to RGB for display 83 | combined_img_rgb = cv2.cvtColor(combined_img, cv2.COLOR_BGR2RGB) 84 | return combined_img_rgb 85 | 86 | 87 | with gr.Blocks(title="Skin Grinding") as iface: 88 | gr.Markdown("## Skin Grinding Application") 89 | 90 | with gr.Row(): 91 | image_input = gr.Image(type="numpy", label="Input Image") 92 | image_output = gr.Image(label="Output Image") 93 | 94 | grind_degree_slider = gr.Slider( 95 | minimum=1, maximum=10, value=3, step=1, label="Grind Degree" 96 | ) 97 | detail_degree_slider = gr.Slider( 98 | minimum=1, maximum=10, value=1, step=1, label="Detail Degree" 99 | ) 100 | strength_slider = gr.Slider( 101 | minimum=0, maximum=10, value=9, step=1, label="Strength" 102 | ) 103 | 104 | gr.Button("Process Image").click( 105 | fn=process_image, 106 | inputs=[ 107 | image_input, 108 | grind_degree_slider, 109 | detail_degree_slider, 110 | strength_slider, 111 | ], 112 | outputs=image_output, 113 | ) 114 | 115 | if __name__ == "__main__": 116 | iface.launch() 117 | -------------------------------------------------------------------------------- /hivision/plugin/beauty/handler.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | from hivision.creator.context import Context 3 | from hivision.plugin.beauty.whitening import make_whitening 4 | from hivision.plugin.beauty.base_adjust import ( 5 | adjust_brightness_contrast_sharpen_saturation, 6 | ) 7 | 8 | 9 | def beauty_face(ctx: Context): 10 | """ 11 | 对人脸进行美颜处理 12 | 1. 美白 13 | 2. 亮度 14 | 15 | :param ctx: Context对象，包含处理参数和图像 16 | """ 17 | middle_image = ctx.origin_image.copy() 18 | processed = False 19 | 20 | # 如果美白强度大于0，进行美白处理 21 | if ctx.params.whitening_strength > 0: 22 | middle_image = make_whitening(middle_image, ctx.params.whitening_strength) 23 | processed = True 24 | 25 | # 如果亮度、对比度、锐化强度不为0，进行亮度、对比度、锐化处理 26 | if ( 27 | ctx.params.brightness_strength != 0 28 | or ctx.params.contrast_strength != 0 29 | or ctx.params.sharpen_strength != 0 30 | or ctx.params.saturation_strength != 0 31 | ): 32 | middle_image = adjust_brightness_contrast_sharpen_saturation( 33 | middle_image, 34 | ctx.params.brightness_strength, 35 | ctx.params.contrast_strength, 36 | ctx.params.sharpen_strength, 37 | ctx.params.saturation_strength, 38 | ) 39 | processed = True 40 | 41 | # 如果进行了美颜处理，更新matting_image 42 | if processed: 43 | # 分离中间图像的BGR通道 44 | b, g, r = cv2.split(middle_image) 45 | # 从原始matting_image中获取alpha通道 46 | _, _, _, alpha = cv2.split(ctx.matting_image) 47 | # 合并处理后的BGR通道和原始alpha通道 48 | ctx.matting_image = cv2.merge((b, g, r, alpha)) 49 | -------------------------------------------------------------------------------- /hivision/plugin/beauty/lut/lut_origin.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/hivision/plugin/beauty/lut/lut_origin.png -------------------------------------------------------------------------------- /hivision/plugin/beauty/thin_face.py: -------------------------------------------------------------------------------- 1 | """ 2 | @author: cuny 3 | @file: ThinFace.py 4 | @time: 2022/7/2 15:50 5 | @description: 6 | 瘦脸算法，用到了图像局部平移法 7 | 先使用人脸关键点检测，然后再使用图像局部平移法 8 | 需要注意的是，这部分不会包含dlib人脸关键点检测，因为考虑到模型载入的问题 9 | """ 10 | 11 | import cv2 12 | import math 13 | import numpy as np 14 | 15 | 16 | class TranslationWarp(object): 17 | """ 18 | 本类包含瘦脸算法，由于瘦脸算法包含了很多个版本，所以以类的方式呈现 19 | 前两个算法没什么好讲的，网上资料很多 20 | 第三个采用numpy内部的自定义函数处理，在处理速度上有一些提升 21 | 最后采用cv2.map算法，处理速度大幅度提升 22 | """ 23 | 24 | # 瘦脸 25 | @staticmethod 26 | def localTranslationWarp(srcImg, startX, startY, endX, endY, radius): 27 | # 双线性插值法 28 | def BilinearInsert(src, ux, uy): 29 | w, h, c = src.shape 30 | if c == 3: 31 | x1 = int(ux) 32 | x2 = x1 + 1 33 | y1 = int(uy) 34 | y2 = y1 + 1 35 | part1 = ( 36 | src[y1, x1].astype(np.float64) * (float(x2) - ux) * (float(y2) - uy) 37 | ) 38 | part2 = ( 39 | src[y1, x2].astype(np.float64) * (ux - float(x1)) * (float(y2) - uy) 40 | ) 41 | part3 = ( 42 | src[y2, x1].astype(np.float64) * (float(x2) - ux) * (uy - float(y1)) 43 | ) 44 | part4 = ( 45 | src[y2, x2].astype(np.float64) * (ux - float(x1)) * (uy - float(y1)) 46 | ) 47 | insertValue = part1 + part2 + part3 + part4 48 | return insertValue.astype(np.int8) 49 | 50 | ddradius = float(radius * radius) # 圆的半径 51 | copyImg = srcImg.copy() # copy后的图像矩阵 52 | # 计算公式中的|m-c|^2 53 | ddmc = (endX - startX) * (endX - startX) + (endY - startY) * (endY - startY) 54 | H, W, C = srcImg.shape # 获取图像的形状 55 | for i in range(W): 56 | for j in range(H): 57 | # # 计算该点是否在形变圆的范围之内 58 | # # 优化，第一步，直接判断是会在（startX,startY)的矩阵框中 59 | if math.fabs(i - startX) > radius and math.fabs(j - startY) > radius: 60 | continue 61 | distance = (i - startX) * (i - startX) + (j - startY) * (j - startY) 62 | if distance < ddradius: 63 | # 计算出（i,j）坐标的原坐标 64 | # 计算公式中右边平方号里的部分 65 | ratio = (ddradius - distance) / (ddradius - distance + ddmc) 66 | ratio = ratio * ratio 67 | # 映射原位置 68 | UX = i - ratio * (endX - startX) 69 | UY = j - ratio * (endY - startY) 70 | 71 | # 根据双线性插值法得到UX，UY的值 72 | # start_ = time.time() 73 | value = BilinearInsert(srcImg, UX, UY) 74 | # print(f"双线性插值耗时;{time.time() - start_}") 75 | # 改变当前 i ，j的值 76 | copyImg[j, i] = value 77 | return copyImg 78 | 79 | # 瘦脸pro1, 限制了for循环的遍历次数 80 | @staticmethod 81 | def localTranslationWarpLimitFor( 82 | srcImg, startP: np.matrix, endP: np.matrix, radius: float 83 | ): 84 | startX, startY = startP[0, 0], startP[0, 1] 85 | endX, endY = endP[0, 0], endP[0, 1] 86 | 87 | # 双线性插值法 88 | def BilinearInsert(src, ux, uy): 89 | w, h, c = src.shape 90 | if c == 3: 91 | x1 = int(ux) 92 | x2 = x1 + 1 93 | y1 = int(uy) 94 | y2 = y1 + 1 95 | part1 = ( 96 | src[y1, x1].astype(np.float64) * (float(x2) - ux) * (float(y2) - uy) 97 | ) 98 | part2 = ( 99 | src[y1, x2].astype(np.float64) * (ux - float(x1)) * (float(y2) - uy) 100 | ) 101 | part3 = ( 102 | src[y2, x1].astype(np.float64) * (float(x2) - ux) * (uy - float(y1)) 103 | ) 104 | part4 = ( 105 | src[y2, x2].astype(np.float64) * (ux - float(x1)) * (uy - float(y1)) 106 | ) 107 | insertValue = part1 + part2 + part3 + part4 108 | return insertValue.astype(np.int8) 109 | 110 | ddradius = float(radius * radius) # 圆的半径 111 | copyImg = srcImg.copy() # copy后的图像矩阵 112 | # 计算公式中的|m-c|^2 113 | ddmc = (endX - startX) ** 2 + (endY - startY) ** 2 114 | # 计算正方形的左上角起始点 115 | startTX, startTY = ( 116 | startX - math.floor(radius + 1), 117 | startY - math.floor((radius + 1)), 118 | ) 119 | # 计算正方形的右下角的结束点 120 | endTX, endTY = ( 121 | startX + math.floor(radius + 1), 122 | startY + math.floor((radius + 1)), 123 | ) 124 | # 剪切srcImg 125 | srcImg = srcImg[startTY : endTY + 1, startTX : endTX + 1, :] 126 | # db.cv_show(srcImg) 127 | # 裁剪后的图像相当于在x,y都减少了startX - math.floor(radius + 1) 128 | # 原本的endX, endY在切后的坐标点 129 | endX, endY = ( 130 | endX - startX + math.floor(radius + 1), 131 | endY - startY + math.floor(radius + 1), 132 | ) 133 | # 原本的startX, startY剪切后的坐标点 134 | startX, startY = (math.floor(radius + 1), math.floor(radius + 1)) 135 | H, W, C = srcImg.shape # 获取图像的形状 136 | for i in range(W): 137 | for j in range(H): 138 | # 计算该点是否在形变圆的范围之内 139 | # 优化，第一步，直接判断是会在（startX,startY)的矩阵框中 140 | # if math.fabs(i - startX) > radius and math.fabs(j - startY) > radius: 141 | # continue 142 | distance = (i - startX) * (i - startX) + (j - startY) * (j - startY) 143 | if distance < ddradius: 144 | # 计算出（i,j）坐标的原坐标 145 | # 计算公式中右边平方号里的部分 146 | ratio = (ddradius - distance) / (ddradius - distance + ddmc) 147 | ratio = ratio * ratio 148 | # 映射原位置 149 | UX = i - ratio * (endX - startX) 150 | UY = j - ratio * (endY - startY) 151 | 152 | # 根据双线性插值法得到UX，UY的值 153 | # start_ = time.time() 154 | value = BilinearInsert(srcImg, UX, UY) 155 | # print(f"双线性插值耗时;{time.time() - start_}") 156 | # 改变当前 i ，j的值 157 | copyImg[j + startTY, i + startTX] = value 158 | return copyImg 159 | 160 | # # 瘦脸pro2,采用了numpy自定义函数做处理 161 | # def localTranslationWarpNumpy(self, srcImg, startP: np.matrix, endP: np.matrix, radius: float): 162 | # startX , startY = startP[0, 0], startP[0, 1] 163 | # endX, endY = endP[0, 0], endP[0, 1] 164 | # ddradius = float(radius * radius) # 圆的半径 165 | # copyImg = srcImg.copy() # copy后的图像矩阵 166 | # # 计算公式中的|m-c|^2 167 | # ddmc = (endX - startX)**2 + (endY - startY)**2 168 | # # 计算正方形的左上角起始点 169 | # startTX, startTY = (startX - math.floor(radius + 1), startY - math.floor((radius + 1))) 170 | # # 计算正方形的右下角的结束点 171 | # endTX, endTY = (startX + math.floor(radius + 1), startY + math.floor((radius + 1))) 172 | # # 剪切srcImg 173 | # self.thinImage = srcImg[startTY : endTY + 1, startTX : endTX + 1, :] 174 | # # s = self.thinImage 175 | # # db.cv_show(srcImg) 176 | # # 裁剪后的图像相当于在x,y都减少了startX - math.floor(radius + 1) 177 | # # 原本的endX, endY在切后的坐标点 178 | # endX, endY = (endX - startX + math.floor(radius + 1), endY - startY + math.floor(radius + 1)) 179 | # # 原本的startX, startY剪切后的坐标点 180 | # startX ,startY = (math.floor(radius + 1), math.floor(radius + 1)) 181 | # H, W, C = self.thinImage.shape # 获取图像的形状 182 | # index_m = np.arange(H * W).reshape((H, W)) 183 | # triangle_ufunc = np.frompyfunc(self.process, 9, 3) 184 | # # start_ = time.time() 185 | # finalImgB, finalImgG, finalImgR = triangle_ufunc(index_m, self, W, ddradius, ddmc, startX, startY, endX, endY) 186 | # finaleImg = np.dstack((finalImgB, finalImgG, finalImgR)).astype(np.uint8) 187 | # finaleImg = np.fliplr(np.rot90(finaleImg, -1)) 188 | # copyImg[startTY: endTY + 1, startTX: endTX + 1, :] = finaleImg 189 | # # print(f"图像处理耗时;{time.time() - start_}") 190 | # # db.cv_show(copyImg) 191 | # return copyImg 192 | 193 | # 瘦脸pro3,采用opencv内置函数 194 | @staticmethod 195 | def localTranslationWarpFastWithStrength( 196 | srcImg, startP: np.matrix, endP: np.matrix, radius, strength: float = 100.0 197 | ): 198 | """ 199 | 采用opencv内置函数 200 | Args: 201 | srcImg: 源图像 202 | startP: 起点位置 203 | endP: 终点位置 204 | radius: 处理半径 205 | strength: 瘦脸强度，一般取100以上 206 | 207 | Returns: 208 | 209 | """ 210 | startX, startY = startP[0, 0], startP[0, 1] 211 | endX, endY = endP[0, 0], endP[0, 1] 212 | ddradius = float(radius * radius) 213 | # copyImg = np.zeros(srcImg.shape, np.uint8) 214 | # copyImg = srcImg.copy() 215 | 216 | maskImg = np.zeros(srcImg.shape[:2], np.uint8) 217 | cv2.circle(maskImg, (startX, startY), math.ceil(radius), (255, 255, 255), -1) 218 | 219 | K0 = 100 / strength 220 | 221 | # 计算公式中的|m-c|^2 222 | ddmc_x = (endX - startX) * (endX - startX) 223 | ddmc_y = (endY - startY) * (endY - startY) 224 | H, W, C = srcImg.shape 225 | 226 | mapX = np.vstack([np.arange(W).astype(np.float32).reshape(1, -1)] * H) 227 | mapY = np.hstack([np.arange(H).astype(np.float32).reshape(-1, 1)] * W) 228 | 229 | distance_x = (mapX - startX) * (mapX - startX) 230 | distance_y = (mapY - startY) * (mapY - startY) 231 | distance = distance_x + distance_y 232 | K1 = np.sqrt(distance) 233 | ratio_x = (ddradius - distance_x) / (ddradius - distance_x + K0 * ddmc_x) 234 | ratio_y = (ddradius - distance_y) / (ddradius - distance_y + K0 * ddmc_y) 235 | ratio_x = ratio_x * ratio_x 236 | ratio_y = ratio_y * ratio_y 237 | 238 | UX = mapX - ratio_x * (endX - startX) * (1 - K1 / radius) 239 | UY = mapY - ratio_y * (endY - startY) * (1 - K1 / radius) 240 | 241 | np.copyto(UX, mapX, where=maskImg == 0) 242 | np.copyto(UY, mapY, where=maskImg == 0) 243 | UX = UX.astype(np.float32) 244 | UY = UY.astype(np.float32) 245 | copyImg = cv2.remap(srcImg, UX, UY, interpolation=cv2.INTER_LINEAR) 246 | return copyImg 247 | 248 | 249 | def thinFace(src, landmark, place: int = 0, strength=30.0): 250 | """ 251 | 瘦脸程序接口，输入人脸关键点信息和强度，即可实现瘦脸 252 | 注意处理四通道图像 253 | Args: 254 | src: 原图 255 | landmark: 关键点信息 256 | place: 选择瘦脸区域，为0-4之间的值 257 | strength: 瘦脸强度，输入值在0-10之间，如果小于或者等于0，则不瘦脸 258 | 259 | Returns: 260 | 瘦脸后的图像 261 | """ 262 | strength = min(100.0, strength * 10.0) 263 | if strength <= 0.0: 264 | return src 265 | # 也可以设置瘦脸区域 266 | place = max(0, min(4, int(place))) 267 | left_landmark = landmark[4 + place] 268 | left_landmark_down = landmark[6 + place] 269 | right_landmark = landmark[13 + place] 270 | right_landmark_down = landmark[15 + place] 271 | endPt = landmark[58] 272 | # 计算第4个点到第6个点的距离作为瘦脸距离 273 | r_left = math.sqrt( 274 | (left_landmark[0, 0] - left_landmark_down[0, 0]) ** 2 275 | + (left_landmark[0, 1] - left_landmark_down[0, 1]) ** 2 276 | ) 277 | 278 | # 计算第14个点到第16个点的距离作为瘦脸距离 279 | r_right = math.sqrt( 280 | (right_landmark[0, 0] - right_landmark_down[0, 0]) ** 2 281 | + (right_landmark[0, 1] - right_landmark_down[0, 1]) ** 2 282 | ) 283 | # 瘦左边脸 284 | thin_image = TranslationWarp.localTranslationWarpFastWithStrength( 285 | src, left_landmark[0], endPt[0], r_left, strength 286 | ) 287 | # 瘦右边脸 288 | thin_image = TranslationWarp.localTranslationWarpFastWithStrength( 289 | thin_image, right_landmark[0], endPt[0], r_right, strength 290 | ) 291 | return thin_image 292 | 293 | 294 | # if __name__ == "__main__": 295 | # import os 296 | # from hycv.FaceDetection68.faceDetection68 import FaceDetection68 297 | 298 | # local_file = os.path.dirname(__file__) 299 | # PREDICTOR_PATH = f"{local_file}/weights/shape_predictor_68_face_landmarks.dat" # 关键点检测模型路径 300 | # fd68 = FaceDetection68(model_path=PREDICTOR_PATH) 301 | # input_image = cv2.imread("test_image/4.jpg", -1) 302 | # _, landmark_, _ = fd68.facePoints(input_image) 303 | # output_image = thinFace(input_image, landmark_, strength=30.2) 304 | # cv2.imwrite("thinFaceCompare.png", np.hstack((input_image, output_image))) 305 | -------------------------------------------------------------------------------- /hivision/plugin/beauty/whitening.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | import numpy as np 3 | import os 4 | import gradio as gr 5 | 6 | 7 | class LutWhite: 8 | CUBE64_ROWS = 8 9 | CUBE64_SIZE = 64 10 | CUBE256_SIZE = 256 11 | CUBE_SCALE = CUBE256_SIZE // CUBE64_SIZE 12 | 13 | def __init__(self, lut_image): 14 | self.lut = self._create_lut(lut_image) 15 | 16 | def _create_lut(self, lut_image): 17 | reshape_lut = np.zeros( 18 | (self.CUBE256_SIZE, self.CUBE256_SIZE, self.CUBE256_SIZE, 3), dtype=np.uint8 19 | ) 20 | for i in range(self.CUBE64_SIZE): 21 | tmp = i // self.CUBE64_ROWS 22 | cx = (i % self.CUBE64_ROWS) * self.CUBE64_SIZE 23 | cy = tmp * self.CUBE64_SIZE 24 | cube64 = lut_image[cy : cy + self.CUBE64_SIZE, cx : cx + self.CUBE64_SIZE] 25 | if cube64.size == 0: 26 | continue 27 | cube256 = cv2.resize(cube64, (self.CUBE256_SIZE, self.CUBE256_SIZE)) 28 | reshape_lut[i * self.CUBE_SCALE : (i + 1) * self.CUBE_SCALE] = cube256 29 | return reshape_lut 30 | 31 | def apply(self, src): 32 | b, g, r = src[:, :, 0], src[:, :, 1], src[:, :, 2] 33 | return self.lut[b, g, r] 34 | 35 | 36 | class MakeWhiter: 37 | def __init__(self, lut_image): 38 | self.lut_white = LutWhite(lut_image) 39 | 40 | def run(self, src: np.ndarray, strength: int) -> np.ndarray: 41 | strength = np.clip(strength / 10.0, 0, 1) 42 | if strength <= 0: 43 | return src 44 | img = self.lut_white.apply(src[:, :, :3]) 45 | return cv2.addWeighted(src[:, :, :3], 1 - strength, img, strength, 0) 46 | 47 | 48 | base_dir = os.path.dirname(os.path.abspath(__file__)) 49 | default_lut = cv2.imread(os.path.join(base_dir, "lut/lut_origin.png")) 50 | make_whiter = MakeWhiter(default_lut) 51 | 52 | 53 | def make_whitening(image, strength): 54 | image = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR) 55 | 56 | iteration = strength // 10 57 | bias = strength % 10 58 | 59 | for i in range(iteration): 60 | image = make_whiter.run(image, 10) 61 | 62 | image = make_whiter.run(image, bias) 63 | 64 | return cv2.cvtColor(image, cv2.COLOR_BGR2RGB) 65 | 66 | 67 | def make_whitening_png(image, strength): 68 | image = cv2.cvtColor(np.array(image), cv2.COLOR_RGBA2BGRA) 69 | 70 | b, g, r, a = cv2.split(image) 71 | bgr_image = cv2.merge((b, g, r)) 72 | 73 | b_w, g_w, r_w = cv2.split(make_whiter.run(bgr_image, strength)) 74 | output_image = cv2.merge((b_w, g_w, r_w, a)) 75 | 76 | return cv2.cvtColor(output_image, cv2.COLOR_RGBA2BGRA) 77 | 78 | 79 | # 启动Gradio应用 80 | if __name__ == "__main__": 81 | demo = gr.Interface( 82 | fn=make_whitening, 83 | inputs=[ 84 | gr.Image(type="pil", image_mode="RGBA", label="Input Image"), 85 | gr.Slider(0, 30, step=1, label="Whitening Strength"), 86 | ], 87 | outputs=gr.Image(type="pil"), 88 | title="Image Whitening Demo", 89 | description="Upload an image and adjust the whitening strength to see the effect.", 90 | ) 91 | demo.launch() 92 | -------------------------------------------------------------------------------- /hivision/plugin/font/.gitkeep: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/hivision/plugin/font/.gitkeep -------------------------------------------------------------------------------- /hivision/plugin/font/青鸟华光简琥珀.ttf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/hivision/plugin/font/青鸟华光简琥珀.ttf -------------------------------------------------------------------------------- /hivision/plugin/template/assets/template_1.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/hivision/plugin/template/assets/template_1.png -------------------------------------------------------------------------------- /hivision/plugin/template/assets/template_2.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zeyi-Lin/HivisionIDPhotos/d993cfb1d8453383254db6cbce2bab8173ac3ae0/hivision/plugin/template/assets/template_2.png -------------------------------------------------------------------------------- /hivision/plugin/template/assets/template_config.json: -------------------------------------------------------------------------------- 1 | { 2 | "template_1": { 3 | "width": 1080, 4 | "height": 1400, 5 | "anchor_points": { 6 | "left_top": [358, 153], 7 | "right_top": [1017, 353], 8 | "left_bottom": [56, 1134], 9 | "right_bottom": [747, 1332], 10 | "rotation": -16.42 11 | } 12 | }, 13 | "template_2": { 14 | "width": 1080, 15 | "height": 1440, 16 | "anchor_points": { 17 | "left_top": [199, 199], 18 | "right_top": [921, 216], 19 | "left_bottom": [163, 1129], 20 | "right_bottom": [876, 1153], 21 | "rotation": -2.2 22 | } 23 | } 24 | } -------------------------------------------------------------------------------- /hivision/plugin/template/template_calculator.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | import numpy as np 3 | import json 4 | from hivision.creator.rotation_adjust import rotate_bound 5 | import os 6 | 7 | base_path = os.path.dirname(os.path.abspath(__file__)) 8 | template_config_path = os.path.join(base_path, 'assets', 'template_config.json') 9 | 10 | def generte_template_photo(template_name: str, input_image: np.ndarray) -> np.ndarray: 11 | """ 12 | 生成模板照片 13 | :param template_name: 模板名称 14 | :param input_image: 输入图像 15 | :return: 模板照片 16 | """ 17 | # 读取模板配置json 18 | with open(template_config_path, 'r') as f: 19 | template_config_dict = json.load(f) 20 | # 获取对应该模板的配置 21 | template_config = template_config_dict[template_name] 22 | 23 | template_width = template_config['width'] 24 | template_height = template_config['height'] 25 | 26 | anchor_points = template_config['anchor_points'] 27 | rotation = anchor_points['rotation'] 28 | left_top = anchor_points['left_top'] 29 | right_top = anchor_points['right_top'] 30 | left_bottom = anchor_points['left_bottom'] 31 | right_bottom = anchor_points['right_bottom'] 32 | 33 | if rotation < 0: 34 | height = right_bottom[1] - left_top[1] 35 | width = right_top[0] - left_bottom[0] 36 | else: 37 | height = left_top[1] - right_bottom[1] 38 | width = left_bottom[0] - right_top[0] 39 | 40 | # 读取模板图像 41 | template_image_path = os.path.join(base_path, 'assets', f'{template_name}.png') 42 | template_image = cv2.imread(template_image_path, cv2.IMREAD_UNCHANGED) 43 | 44 | # 无损旋转 45 | rotated_image = rotate_bound(input_image, -1 * rotation)[0] 46 | rotated_image_height, rotated_image_width, _ = rotated_image.shape 47 | 48 | # 计算缩放比例 49 | scale_x = width / rotated_image_width 50 | scale_y = height / rotated_image_height 51 | scale = max(scale_x, scale_y) 52 | 53 | resized_image = cv2.resize(rotated_image, None, fx=scale, fy=scale) 54 | resized_height, resized_width, _ = resized_image.shape 55 | 56 | # 创建一个与template_image大小相同的背景，使用白色填充 57 | result = np.full((template_height, template_width, 3), 255, dtype=np.uint8) 58 | 59 | # 计算粘贴位置 60 | paste_x = left_bottom[0] 61 | paste_y = left_top[1] 62 | 63 | # 确保不会超出边界 64 | paste_height = min(resized_height, template_height - paste_y) 65 | paste_width = min(resized_width, template_width - paste_x) 66 | 67 | # 将旋转后的图像粘贴到结果图像上 68 | result[paste_y:paste_y+paste_height, paste_x:paste_x+paste_width] = resized_image[:paste_height, :paste_width] 69 | 70 | template_image = cv2.cvtColor(template_image, cv2.COLOR_BGRA2RGBA) 71 | 72 | # 将template_image叠加到结果图像上 73 | if template_image.shape[2] == 4: # 确保template_image有alpha通道 74 | alpha = template_image[:, :, 3] / 255.0 75 | for c in range(0, 3): 76 | result[:, :, c] = result[:, :, c] * (1 - alpha) + template_image[:, :, c] * alpha 77 | 78 | return result 79 | -------------------------------------------------------------------------------- /hivision/plugin/watermark.py: -------------------------------------------------------------------------------- 1 | """ 2 | Reference: https://gist.github.com/Deali-Axy/e22ea79bfbe785f9017b2e3cd7fdb3eb 3 | """ 4 | 5 | import enum 6 | import os 7 | import math 8 | import textwrap 9 | from PIL import Image, ImageFont, ImageDraw, ImageEnhance, ImageChops 10 | import os 11 | 12 | base_path = os.path.abspath(os.path.dirname(__file__)) 13 | 14 | 15 | class WatermarkerStyles(enum.Enum): 16 | """水印样式""" 17 | 18 | STRIPED = 1 # 斜向重复 19 | CENTRAL = 2 # 居中 20 | 21 | 22 | class Watermarker(object): 23 | """图片水印工具""" 24 | 25 | def __init__( 26 | self, 27 | input_image: Image.Image, 28 | text: str, 29 | style: WatermarkerStyles, 30 | angle=30, 31 | color="#8B8B1B", 32 | font_file="青鸟华光简琥珀.ttf", 33 | opacity=0.15, 34 | size=50, 35 | space=75, 36 | chars_per_line=8, 37 | font_height_crop=1.2, 38 | ): 39 | """_summary_ 40 | 41 | Parameters 42 | ---------- 43 | input_image : Image.Image 44 | PIL图片对象 45 | text : str 46 | 水印文字 47 | style : WatermarkerStyles 48 | 水印样式 49 | angle : int, optional 50 | 水印角度, by default 30 51 | color : str, optional 52 | 水印颜色, by default "#8B8B1B" 53 | font_file : str, optional 54 | 字体文件, by default "青鸟华光简琥珀.ttf" 55 | font_height_crop : float, optional 56 | 字体高度裁剪比例, by default 1.2 57 | opacity : float, optional 58 | 水印透明度, by default 0.15 59 | size : int, optional 60 | 字体大小, by default 50 61 | space : int, optional 62 | 水印间距, by default 75 63 | chars_per_line : int, optional 64 | 每行字符数, by default 8 65 | """ 66 | self.input_image = input_image 67 | self.text = text 68 | self.style = style 69 | self.angle = angle 70 | self.color = color 71 | self.font_file = os.path.join(base_path, "font", font_file) 72 | self.font_height_crop = font_height_crop 73 | self.opacity = opacity 74 | self.size = size 75 | self.space = space 76 | self.chars_per_line = chars_per_line 77 | self._result_image = None 78 | 79 | @staticmethod 80 | def set_image_opacity(image: Image, opacity: float): 81 | alpha = image.split()[3] 82 | alpha = ImageEnhance.Brightness(alpha).enhance(opacity) 83 | image.putalpha(alpha) 84 | return image 85 | 86 | @staticmethod 87 | def crop_image_edge(image: Image): 88 | bg = Image.new(mode="RGBA", size=image.size) 89 | diff = ImageChops.difference(image, bg) 90 | bbox = diff.getbbox() 91 | if bbox: 92 | return image.crop(bbox) 93 | return image 94 | 95 | def _add_mark_striped(self): 96 | origin_image = self.input_image.convert("RGBA") 97 | width = len(self.text) * self.size 98 | height = round(self.size * self.font_height_crop) 99 | watermark_image = Image.new(mode="RGBA", size=(width, height)) 100 | draw_table = ImageDraw.Draw(watermark_image) 101 | draw_table.text( 102 | (0, 0), 103 | self.text, 104 | fill=self.color, 105 | font=ImageFont.truetype(self.font_file, size=self.size), 106 | ) 107 | watermark_image = Watermarker.crop_image_edge(watermark_image) 108 | Watermarker.set_image_opacity(watermark_image, self.opacity) 109 | 110 | c = int(math.sqrt(origin_image.size[0] ** 2 + origin_image.size[1] ** 2)) 111 | watermark_mask = Image.new(mode="RGBA", size=(c, c)) 112 | y, idx = 0, 0 113 | while y < c: 114 | x = -int((watermark_image.size[0] + self.space) * 0.5 * idx) 115 | idx = (idx + 1) % 2 116 | while x < c: 117 | watermark_mask.paste(watermark_image, (x, y)) 118 | x += watermark_image.size[0] + self.space 119 | y += watermark_image.size[1] + self.space 120 | 121 | watermark_mask = watermark_mask.rotate(self.angle) 122 | origin_image.paste( 123 | watermark_mask, 124 | (int((origin_image.size[0] - c) / 2), int((origin_image.size[1] - c) / 2)), 125 | mask=watermark_mask.split()[3], 126 | ) 127 | return origin_image 128 | 129 | def _add_mark_central(self): 130 | origin_image = self.input_image.convert("RGBA") 131 | text_lines = textwrap.wrap(self.text, width=self.chars_per_line) 132 | text = "\n".join(text_lines) 133 | width = len(text) * self.size 134 | height = round(self.size * self.font_height_crop * len(text_lines)) 135 | watermark_image = Image.new(mode="RGBA", size=(width, height)) 136 | draw_table = ImageDraw.Draw(watermark_image) 137 | draw_table.text( 138 | (0, 0), 139 | text, 140 | fill=self.color, 141 | font=ImageFont.truetype(self.font_file, size=self.size), 142 | ) 143 | watermark_image = Watermarker.crop_image_edge(watermark_image) 144 | Watermarker.set_image_opacity(watermark_image, self.opacity) 145 | 146 | c = int(math.sqrt(origin_image.size[0] ** 2 + origin_image.size[1] ** 2)) 147 | watermark_mask = Image.new(mode="RGBA", size=(c, c)) 148 | watermark_mask.paste( 149 | watermark_image, 150 | ( 151 | int((watermark_mask.width - watermark_image.width) / 2), 152 | int((watermark_mask.height - watermark_image.height) / 2), 153 | ), 154 | ) 155 | watermark_mask = watermark_mask.rotate(self.angle) 156 | 157 | origin_image.paste( 158 | watermark_mask, 159 | ( 160 | int((origin_image.width - watermark_mask.width) / 2), 161 | int((origin_image.height - watermark_mask.height) / 2), 162 | ), 163 | mask=watermark_mask.split()[3], 164 | ) 165 | return origin_image 166 | 167 | @property 168 | def image(self): 169 | if not self._result_image: 170 | if self.style == WatermarkerStyles.STRIPED: 171 | self._result_image = self._add_mark_striped() 172 | elif self.style == WatermarkerStyles.CENTRAL: 173 | self._result_image = self._add_mark_central() 174 | return self._result_image 175 | 176 | def save(self, file_path: str, image_format: str = "png"): 177 | with open(file_path, "wb") as f: 178 | self.image.save(f, image_format) 179 | 180 | 181 | # Gradio 接口 182 | def watermark_image( 183 | image, 184 | text, 185 | style, 186 | angle, 187 | color, 188 | opacity, 189 | size, 190 | space, 191 | ): 192 | # 创建 Watermarker 实例 193 | watermarker = Watermarker( 194 | input_image=image, 195 | text=text, 196 | style=( 197 | WatermarkerStyles.STRIPED 198 | if style == "STRIPED" 199 | else WatermarkerStyles.CENTRAL 200 | ), 201 | angle=angle, 202 | color=color, 203 | opacity=opacity, 204 | size=size, 205 | space=space, 206 | ) 207 | 208 | # 返回带水印的图片 209 | return watermarker.image 210 | 211 | 212 | if __name__ == "__main__": 213 | import gradio as gr 214 | 215 | iface = gr.Interface( 216 | fn=watermark_image, 217 | inputs=[ 218 | gr.Image(type="pil", label="上传图片", height=400), 219 | gr.Textbox(label="水印文字"), 220 | gr.Radio(choices=["STRIPED", "CENTRAL"], label="水印样式"), 221 | gr.Slider(minimum=0, maximum=360, value=30, label="水印角度"), 222 | gr.ColorPicker(label="水印颜色"), 223 | gr.Slider(minimum=0, maximum=1, value=0.15, label="水印透明度"), 224 | gr.Slider(minimum=10, maximum=100, value=50, label="字体大小"), 225 | gr.Slider(minimum=10, maximum=200, value=75, label="水印间距"), 226 | ], 227 | outputs=gr.Image(type="pil", label="带水印的图片", height=400), 228 | title="图片水印工具", 229 | description="上传一张图片，添加水印并下载。", 230 | ) 231 | 232 | iface.launch() 233 | -------------------------------------------------------------------------------- /hivision/utils.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding: utf-8 -*- 3 | from PIL import Image 4 | import io 5 | import numpy as np 6 | import cv2 7 | import base64 8 | from hivision.plugin.watermark import Watermarker, WatermarkerStyles 9 | 10 | 11 | def save_image_dpi_to_bytes(image: np.ndarray, output_image_path: str = None, dpi: int = 300): 12 | """ 13 | 设置图像的DPI（每英寸点数）并返回字节流 14 | 15 | :param image: numpy.ndarray, 输入的图像数组 16 | :param output_image_path: Path to save the resized image. 保存调整大小后的图像的路径。 17 | :param dpi: int, 要设置的DPI值，默认为300 18 | """ 19 | image = Image.fromarray(image) 20 | # 创建一个字节流对象 21 | byte_stream = io.BytesIO() 22 | # 将图像保存到字节流 23 | image.save(byte_stream, format="PNG", dpi=(dpi, dpi)) 24 | # 获取字节流的内容 25 | image_bytes = byte_stream.getvalue() 26 | 27 | # Save the image to the output path 28 | if output_image_path: 29 | with open(output_image_path, "wb") as f: 30 | f.write(image_bytes) 31 | 32 | return image_bytes 33 | 34 | 35 | def resize_image_to_kb(input_image: np.ndarray, output_image_path: str = None, target_size_kb: int = 100, dpi: int = 300): 36 | """ 37 | Resize an image to a target size in KB. 38 | 将图像调整大小至目标文件大小（KB）。 39 | 40 | :param input_image_path: Path to the input image. 输入图像的路径。 41 | :param output_image_path: Path to save the resized image. 保存调整大小后的图像的路径。 42 | :param target_size_kb: Target size in KB. 目标文件大小（KB）。 43 | 44 | Example: 45 | resize_image_to_kb('input_image.jpg', 'output_image.jpg', 50) 46 | """ 47 | 48 | if isinstance(input_image, np.ndarray): 49 | img = Image.fromarray(input_image) 50 | elif isinstance(input_image, Image.Image): 51 | img = input_image 52 | else: 53 | raise ValueError("input_image must be a NumPy array or PIL Image.") 54 | 55 | # Convert image to RGB mode if it's not 56 | if img.mode != "RGB": 57 | img = img.convert("RGB") 58 | 59 | # Initial quality value 60 | quality = 95 61 | 62 | while True: 63 | # Create a BytesIO object to hold the image data in memory 64 | img_byte_arr = io.BytesIO() 65 | 66 | # Save the image to the BytesIO object with the current quality 67 | img.save(img_byte_arr, format="JPEG", quality=quality, dpi=(dpi, dpi)) 68 | 69 | # Get the size of the image in KB 70 | img_size_kb = len(img_byte_arr.getvalue()) / 1024 71 | 72 | # Check if the image size is within the target size 73 | if img_size_kb <= target_size_kb or quality == 1: 74 | # If the image is smaller than the target size, add padding 75 | if img_size_kb < target_size_kb: 76 | padding_size = int( 77 | (target_size_kb * 1024) - len(img_byte_arr.getvalue()) 78 | ) 79 | padding = b"\x00" * padding_size 80 | img_byte_arr.write(padding) 81 | 82 | # Save the image to the output path 83 | if output_image_path: 84 | with open(output_image_path, "wb") as f: 85 | f.write(img_byte_arr.getvalue()) 86 | 87 | return img_byte_arr.getvalue() 88 | 89 | # Reduce the quality if the image is still too large 90 | quality -= 5 91 | 92 | # Ensure quality does not go below 1 93 | if quality < 1: 94 | quality = 1 95 | 96 | 97 | def resize_image_to_kb_base64(input_image, target_size_kb, mode="exact"): 98 | """ 99 | Resize an image to a target size in KB and return it as a base64 encoded string. 100 | 将图像调整大小至目标文件大小（KB）并返回base64编码的字符串。 101 | 102 | :param input_image: Input image as a NumPy array or PIL Image. 输入图像，可以是NumPy数组或PIL图像。 103 | :param target_size_kb: Target size in KB. 目标文件大小（KB）。 104 | :param mode: Mode of resizing ('exact', 'max', 'min'). 模式：'exact'（精确大小）、'max'（不大于）、'min'（不小于）。 105 | 106 | :return: Base64 encoded string of the resized image. 调整大小后的图像的base64编码字符串。 107 | """ 108 | 109 | if isinstance(input_image, np.ndarray): 110 | img = Image.fromarray(input_image) 111 | elif isinstance(input_image, Image.Image): 112 | img = input_image 113 | else: 114 | raise ValueError("input_image must be a NumPy array or PIL Image.") 115 | 116 | # Convert image to RGB mode if it's not 117 | if img.mode != "RGB": 118 | img = img.convert("RGB") 119 | 120 | # Initial quality value 121 | quality = 95 122 | 123 | while True: 124 | # Create a BytesIO object to hold the image data in memory 125 | img_byte_arr = io.BytesIO() 126 | 127 | # Save the image to the BytesIO object with the current quality 128 | img.save(img_byte_arr, format="JPEG", quality=quality) 129 | 130 | # Get the size of the image in KB 131 | img_size_kb = len(img_byte_arr.getvalue()) / 1024 132 | 133 | # Check based on the mode 134 | if mode == "exact": 135 | # If the image size is equal to the target size, we can return it 136 | if img_size_kb == target_size_kb: 137 | break 138 | 139 | # If the image is smaller than the target size, add padding 140 | elif img_size_kb < target_size_kb: 141 | padding_size = int( 142 | (target_size_kb * 1024) - len(img_byte_arr.getvalue()) 143 | ) 144 | padding = b"\x00" * padding_size 145 | img_byte_arr.write(padding) 146 | break 147 | 148 | elif mode == "max": 149 | # If the image size is within the target size, we can return it 150 | if img_size_kb <= target_size_kb or quality == 1: 151 | break 152 | 153 | elif mode == "min": 154 | # If the image size is greater than or equal to the target size, we can return it 155 | if img_size_kb >= target_size_kb: 156 | break 157 | 158 | # Reduce the quality if the image is still too large 159 | quality -= 5 160 | 161 | # Ensure quality does not go below 1 162 | if quality < 1: 163 | quality = 1 164 | 165 | # Encode the image data to base64 166 | img_base64 = base64.b64encode(img_byte_arr.getvalue()).decode("utf-8") 167 | return "data:image/png;base64," + img_base64 168 | 169 | 170 | def numpy_2_base64(img: np.ndarray) -> str: 171 | _, buffer = cv2.imencode(".png", img) 172 | base64_image = base64.b64encode(buffer).decode("utf-8") 173 | 174 | return "data:image/png;base64," + base64_image 175 | 176 | 177 | def base64_2_numpy(base64_image: str) -> np.ndarray: 178 | # Remove the data URL prefix if present 179 | if base64_image.startswith('data:image'): 180 | base64_image = base64_image.split(',')[1] 181 | 182 | # Decode base64 string to bytes 183 | img_bytes = base64.b64decode(base64_image) 184 | 185 | # Convert bytes to numpy array 186 | img_array = np.frombuffer(img_bytes, dtype=np.uint8) 187 | 188 | # Decode the image array 189 | img = cv2.imdecode(img_array, cv2.IMREAD_UNCHANGED) 190 | 191 | return img 192 | 193 | # 字节流转base64 194 | def bytes_2_base64(img_byte_arr: bytes) -> str: 195 | base64_image = base64.b64encode(img_byte_arr).decode("utf-8") 196 | return "data:image/png;base64," + base64_image 197 | 198 | 199 | def save_numpy_image(numpy_img, file_path): 200 | # 检查数组的形状 201 | if numpy_img.shape[2] == 4: 202 | # 将 BGR 转换为 RGB，并保留透明通道 203 | rgb_img = np.concatenate( 204 | (np.flip(numpy_img[:, :, :3], axis=-1), numpy_img[:, :, 3:]), axis=-1 205 | ).astype(np.uint8) 206 | img = Image.fromarray(rgb_img, mode="RGBA") 207 | else: 208 | # 将 BGR 转换为 RGB 209 | rgb_img = np.flip(numpy_img, axis=-1).astype(np.uint8) 210 | img = Image.fromarray(rgb_img, mode="RGB") 211 | 212 | img.save(file_path) 213 | 214 | 215 | def numpy_to_bytes(numpy_img): 216 | img = Image.fromarray(numpy_img) 217 | img_byte_arr = io.BytesIO() 218 | img.save(img_byte_arr, format="PNG") 219 | img_byte_arr.seek(0) 220 | return img_byte_arr 221 | 222 | 223 | def hex_to_rgb(value): 224 | value = value.lstrip("#") 225 | length = len(value) 226 | return tuple( 227 | int(value[i : i + length // 3], 16) for i in range(0, length, length // 3) 228 | ) 229 | 230 | 231 | def generate_gradient(start_color, width, height, mode="updown"): 232 | # 定义背景颜色 233 | end_color = (255, 255, 255) # 白色 234 | 235 | # 创建一个空白图像 236 | r_out = np.zeros((height, width), dtype=int) 237 | g_out = np.zeros((height, width), dtype=int) 238 | b_out = np.zeros((height, width), dtype=int) 239 | 240 | if mode == "updown": 241 | # 生成上下渐变色 242 | for y in range(height): 243 | r = int( 244 | (y / height) * end_color[0] + ((height - y) / height) * start_color[0] 245 | ) 246 | g = int( 247 | (y / height) * end_color[1] + ((height - y) / height) * start_color[1] 248 | ) 249 | b = int( 250 | (y / height) * end_color[2] + ((height - y) / height) * start_color[2] 251 | ) 252 | r_out[y, :] = r 253 | g_out[y, :] = g 254 | b_out[y, :] = b 255 | 256 | else: 257 | # 生成中心渐变色 258 | img = np.zeros((height, width, 3)) 259 | # 定义椭圆中心和半径 260 | center = (width // 2, height // 2) 261 | end_axies = max(height, width) 262 | # 定义渐变色 263 | end_color = (255, 255, 255) 264 | # 绘制椭圆 265 | for y in range(end_axies): 266 | axes = (end_axies - y, end_axies - y) 267 | r = int( 268 | (y / end_axies) * end_color[0] 269 | + ((end_axies - y) / end_axies) * start_color[0] 270 | ) 271 | g = int( 272 | (y / end_axies) * end_color[1] 273 | + ((end_axies - y) / end_axies) * start_color[1] 274 | ) 275 | b = int( 276 | (y / end_axies) * end_color[2] 277 | + ((end_axies - y) / end_axies) * start_color[2] 278 | ) 279 | 280 | cv2.ellipse(img, center, axes, 0, 0, 360, (b, g, r), -1) 281 | b_out, g_out, r_out = cv2.split(np.uint64(img)) 282 | 283 | return r_out, g_out, b_out 284 | 285 | 286 | def add_background(input_image, bgr=(0, 0, 0), mode="pure_color"): 287 | """ 288 | 本函数的功能为为透明图像加上背景。 289 | :param input_image: numpy.array(4 channels), 透明图像 290 | :param bgr: tuple, 合成纯色底时的 BGR 值 291 | :param new_background: numpy.array(3 channels)，合成自定义图像底时的背景图 292 | :return: output: 合成好的输出图像 293 | """ 294 | height, width = input_image.shape[0], input_image.shape[1] 295 | try: 296 | b, g, r, a = cv2.split(input_image) 297 | except ValueError: 298 | raise ValueError( 299 | "The input image must have 4 channels. 输入图像必须有4个通道，即透明图像。" 300 | ) 301 | 302 | a_cal = a / 255 303 | if mode == "pure_color": 304 | # 纯色填充 305 | b2 = np.full([height, width], bgr[0], dtype=int) 306 | g2 = np.full([height, width], bgr[1], dtype=int) 307 | r2 = np.full([height, width], bgr[2], dtype=int) 308 | elif mode == "updown_gradient": 309 | b2, g2, r2 = generate_gradient(bgr, width, height, mode="updown") 310 | else: 311 | b2, g2, r2 = generate_gradient(bgr, width, height, mode="center") 312 | 313 | output = cv2.merge( 314 | ((b - b2) * a_cal + b2, (g - g2) * a_cal + g2, (r - r2) * a_cal + r2) 315 | ) 316 | 317 | return output 318 | 319 | def add_background_with_image(input_image: np.ndarray, background_image: np.ndarray) -> np.ndarray: 320 | """ 321 | 本函数的功能为为透明图像加上背景。 322 | :param input_image: numpy.array(4 channels), 透明图像 323 | :param background_image: numpy.array(3 channels), 背景图像 324 | :return: output: 合成好的输出图像 325 | """ 326 | height, width = input_image.shape[:2] 327 | try: 328 | b, g, r, a = cv2.split(input_image) 329 | except ValueError: 330 | raise ValueError( 331 | "The input image must have 4 channels. 输入图像必须有4个通道，即透明图像。" 332 | ) 333 | 334 | # 确保背景图像与输入图像大小一致 335 | background_image = cv2.resize(background_image, (width, height), cv2.INTER_AREA) 336 | background_image = cv2.cvtColor(background_image, cv2.COLOR_BGR2RGB) 337 | b2, g2, r2 = cv2.split(background_image) 338 | 339 | a_cal = a / 255.0 340 | 341 | # 修正混合公式 342 | output = cv2.merge( 343 | (b * a_cal + b2 * (1 - a_cal), 344 | g * a_cal + g2 * (1 - a_cal), 345 | r * a_cal + r2 * (1 - a_cal)) 346 | ) 347 | 348 | return output.astype(np.uint8) 349 | 350 | def add_watermark( 351 | image, text, size=50, opacity=0.5, angle=45, color="#8B8B1B", space=75 352 | ): 353 | image = Image.fromarray(image) 354 | watermarker = Watermarker( 355 | input_image=image, 356 | text=text, 357 | style=WatermarkerStyles.STRIPED, 358 | angle=angle, 359 | color=color, 360 | opacity=opacity, 361 | size=size, 362 | space=space, 363 | ) 364 | return np.array(watermarker.image.convert("RGB")) 365 | -------------------------------------------------------------------------------- /inference.py: -------------------------------------------------------------------------------- 1 | import os 2 | import cv2 3 | import argparse 4 | import numpy as np 5 | from hivision.error import FaceError 6 | from hivision.utils import hex_to_rgb, resize_image_to_kb, add_background, save_image_dpi_to_bytes 7 | from hivision import IDCreator 8 | from hivision.creator.layout_calculator import ( 9 | generate_layout_array, 10 | generate_layout_image, 11 | ) 12 | from hivision.creator.choose_handler import choose_handler 13 | from hivision.utils import hex_to_rgb, resize_image_to_kb 14 | 15 | 16 | INFERENCE_TYPE = [ 17 | "idphoto", 18 | "human_matting", 19 | "add_background", 20 | "generate_layout_photos", 21 | "idphoto_crop", 22 | ] 23 | MATTING_MODEL = [ 24 | "hivision_modnet", 25 | "modnet_photographic_portrait_matting", 26 | "mnn_hivision_modnet", 27 | "rmbg-1.4", 28 | "birefnet-v1-lite", 29 | ] 30 | FACE_DETECT_MODEL = [ 31 | "mtcnn", 32 | "face_plusplus", 33 | "retinaface-resnet50", 34 | ] 35 | RENDER = [0, 1, 2] 36 | 37 | parser = argparse.ArgumentParser(description="HivisionIDPhotos 证件照制作推理程序。") 38 | parser.add_argument( 39 | "-t", 40 | "--type", 41 | help="请求 API 的种类", 42 | choices=INFERENCE_TYPE, 43 | default="idphoto", 44 | ) 45 | parser.add_argument("-i", "--input_image_dir", help="输入图像路径", required=True) 46 | parser.add_argument("-o", "--output_image_dir", help="保存图像路径", required=True) 47 | parser.add_argument("--height", help="证件照尺寸-高", default=413) 48 | parser.add_argument("--width", help="证件照尺寸-宽", default=295) 49 | parser.add_argument("-c", "--color", help="证件照背景色", default="638cce") 50 | parser.add_argument("--hd", type=bool, help="是否输出高清照", default=True) 51 | parser.add_argument( 52 | "-k", "--kb", help="输出照片的 KB 值，仅对换底和制作排版照生效", default=None 53 | ) 54 | parser.add_argument( 55 | "-r", 56 | "--render", 57 | type=int, 58 | help="底色合成的模式，有 0:纯色、1:上下渐变、2:中心渐变 可选", 59 | choices=RENDER, 60 | default=0, 61 | ) 62 | parser.add_argument( 63 | "--dpi", 64 | type=int, 65 | help="输出照片的 DPI 值", 66 | default=300, 67 | ) 68 | parser.add_argument( 69 | "--face_align", 70 | type=bool, 71 | help="是否进行人脸旋转矫正", 72 | default=False, 73 | ) 74 | parser.add_argument( 75 | "--matting_model", 76 | help="抠图模型权重", 77 | default="modnet_photographic_portrait_matting", 78 | choices=MATTING_MODEL, 79 | ) 80 | parser.add_argument( 81 | "--face_detect_model", 82 | help="人脸检测模型", 83 | default="mtcnn", 84 | choices=FACE_DETECT_MODEL, 85 | ) 86 | 87 | args = parser.parse_args() 88 | 89 | # ------------------- 选择抠图与人脸检测模型 ------------------- 90 | creator = IDCreator() 91 | choose_handler(creator, args.matting_model, args.face_detect_model) 92 | 93 | root_dir = os.path.dirname(os.path.abspath(__file__)) 94 | input_image = cv2.imread(args.input_image_dir, cv2.IMREAD_UNCHANGED) 95 | 96 | # 如果模式是生成证件照 97 | if args.type == "idphoto": 98 | # 将字符串转为元组 99 | size = (int(args.height), int(args.width)) 100 | try: 101 | result = creator(input_image, size=size, face_alignment=args.face_align) 102 | except FaceError: 103 | print("人脸数量不等于 1，请上传单张人脸的图像。") 104 | else: 105 | # 保存标准照 106 | save_image_dpi_to_bytes(cv2.cvtColor(result.standard, cv2.COLOR_RGBA2BGRA), args.output_image_dir, dpi=args.dpi) 107 | 108 | # 保存高清照 109 | file_name, file_extension = os.path.splitext(args.output_image_dir) 110 | new_file_name = file_name + "_hd" + file_extension 111 | save_image_dpi_to_bytes(cv2.cvtColor(result.hd, cv2.COLOR_RGBA2BGRA), new_file_name, dpi=args.dpi) 112 | 113 | # 如果模式是人像抠图 114 | elif args.type == "human_matting": 115 | result = creator(input_image, change_bg_only=True) 116 | cv2.imwrite(args.output_image_dir, result.hd) 117 | 118 | # 如果模式是添加背景 119 | elif args.type == "add_background": 120 | 121 | render_choice = ["pure_color", "updown_gradient", "center_gradient"] 122 | 123 | # 将字符串转为元组 124 | color = hex_to_rgb(args.color) 125 | # 将元祖的 0 和 2 号数字交换 126 | color = (color[2], color[1], color[0]) 127 | 128 | result_image = add_background( 129 | input_image, bgr=color, mode=render_choice[args.render] 130 | ) 131 | result_image = result_image.astype(np.uint8) 132 | result_image = cv2.cvtColor(result_image, cv2.COLOR_RGBA2BGRA) 133 | 134 | if args.kb: 135 | resize_image_to_kb(result_image, args.output_image_dir, int(args.kb), dpi=args.dpi) 136 | else: 137 | save_image_dpi_to_bytes(cv2.cvtColor(result_image, cv2.COLOR_RGBA2BGRA), args.output_image_dir, dpi=args.dpi) 138 | 139 | # 如果模式是生成排版照 140 | elif args.type == "generate_layout_photos": 141 | 142 | size = (int(args.height), int(args.width)) 143 | 144 | typography_arr, typography_rotate = generate_layout_array( 145 | input_height=size[0], input_width=size[1] 146 | ) 147 | 148 | result_layout_image = generate_layout_image( 149 | input_image, 150 | typography_arr, 151 | typography_rotate, 152 | height=size[0], 153 | width=size[1], 154 | ) 155 | 156 | if args.kb: 157 | result_layout_image = cv2.cvtColor(result_layout_image, cv2.COLOR_RGB2BGR) 158 | result_layout_image = resize_image_to_kb( 159 | result_layout_image, args.output_image_dir, int(args.kb), dpi=args.dpi 160 | ) 161 | else: 162 | save_image_dpi_to_bytes(cv2.cvtColor(result_layout_image, cv2.COLOR_RGBA2BGRA), args.output_image_dir, dpi=args.dpi) 163 | 164 | # 如果模式是证件照裁切 165 | elif args.type == "idphoto_crop": 166 | # 将字符串转为元组 167 | size = (int(args.height), int(args.width)) 168 | try: 169 | result = creator(input_image, size=size, crop_only=True) 170 | except FaceError: 171 | print("人脸数量不等于 1，请上传单张人脸的图像。") 172 | else: 173 | # 保存标准照 174 | save_image_dpi_to_bytes(cv2.cvtColor(result.standard, cv2.COLOR_RGBA2BGRA), args.output_image_dir, dpi=args.dpi) 175 | 176 | # 保存高清照 177 | file_name, file_extension = os.path.splitext(args.output_image_dir) 178 | new_file_name = file_name + "_hd" + file_extension 179 | save_image_dpi_to_bytes(cv2.cvtColor(result.hd, cv2.COLOR_RGBA2BGRA), new_file_name, dpi=args.dpi) -------------------------------------------------------------------------------- /requirements-app.txt: -------------------------------------------------------------------------------- 1 | gradio>=4.43.0 2 | fastapi -------------------------------------------------------------------------------- /requirements-dev.txt: -------------------------------------------------------------------------------- 1 | black -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | opencv-python>=4.8.1.78 2 | onnxruntime>=1.15.0 3 | numpy<=1.26.4 4 | requests 5 | mtcnn-runtime 6 | tqdm 7 | starlette 8 | -------------------------------------------------------------------------------- /scripts/build_pypi.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding: utf-8 -*- 3 | r""" 4 | @DATE: 2024/9/5 16:56 5 | @File: build_pypi.py 6 | @IDE: pycharm 7 | @Description: 8 | 构建pypi包 9 | """ 10 | -------------------------------------------------------------------------------- /scripts/download_model.py: -------------------------------------------------------------------------------- 1 | import os 2 | import requests 3 | import argparse 4 | from tqdm import tqdm # 导入 tqdm 库 5 | 6 | # 获取当前脚本所在目录的上一级目录 7 | base_path = os.path.dirname(os.path.dirname(os.path.abspath(__file__))) 8 | 9 | 10 | def download_file(url, save_path): 11 | try: 12 | print(f"Begin downloading: {url}") 13 | response = requests.get(url, stream=True) 14 | response.raise_for_status() # 检查请求是否成功 15 | 16 | # 获取文件总大小 17 | total_size = int(response.headers.get("content-length", 0)) 18 | # 使用 tqdm 显示进度条 19 | with open(save_path, "wb") as file, tqdm( 20 | total=total_size, 21 | unit="B", 22 | unit_scale=True, 23 | unit_divisor=1024, 24 | desc=os.path.basename(save_path), 25 | ) as bar: 26 | for chunk in response.iter_content(chunk_size=8192): 27 | file.write(chunk) 28 | bar.update(len(chunk)) # 更新进度条 29 | print(f"Download completed. Save to: {save_path}") 30 | except requests.exceptions.RequestException as e: 31 | print(f"Download failed: {e}") 32 | 33 | 34 | def download_models(model_urls): 35 | # 下载每个模型 36 | for model_name, model_info in model_urls.items(): 37 | # 指定下载保存的目录 38 | save_dir = model_info["location"] 39 | 40 | # 创建目录（如果不存在的话） 41 | os.makedirs(os.path.join(base_path, save_dir), exist_ok=True) 42 | 43 | url = model_info["url"] 44 | file_format = model_info["format"] 45 | 46 | # 特殊处理 rmbg-1.4 模型的文件名 47 | file_name = f"{model_name}.{file_format}" 48 | 49 | save_path = os.path.join(base_path, save_dir, file_name) 50 | 51 | # 检查文件是否已经存在 52 | if os.path.exists(save_path): 53 | print(f"File already exists, skipping download: {save_path}") 54 | continue 55 | 56 | # 下载文件 57 | download_file(url, save_path) 58 | 59 | 60 | def main(models_to_download): 61 | # 模型权重的下载链接 62 | model_urls = { 63 | "hivision_modnet": { 64 | "url": "https://github.com/Zeyi-Lin/HivisionIDPhotos/releases/download/pretrained-model/hivision_modnet.onnx", 65 | "format": "onnx", 66 | "location": "hivision/creator/weights", 67 | }, 68 | "modnet_photographic_portrait_matting": { 69 | "url": "https://github.com/Zeyi-Lin/HivisionIDPhotos/releases/download/pretrained-model/modnet_photographic_portrait_matting.onnx", 70 | "format": "onnx", 71 | "location": "hivision/creator/weights", 72 | }, 73 | # "mnn_hivision_modnet": { 74 | # "url": "https://github.com/Zeyi-Lin/HivisionIDPhotos/releases/download/pretrained-model/mnn_hivision_modnet.mnn", 75 | # "format": "mnn", 76 | # }, 77 | "rmbg-1.4": { 78 | "url": "https://huggingface.co/briaai/RMBG-1.4/resolve/main/onnx/model.onnx?download=true", 79 | "format": "onnx", 80 | "location": "hivision/creator/weights", 81 | }, 82 | "birefnet-v1-lite": { 83 | "url": "https://github.com/ZhengPeng7/BiRefNet/releases/download/v1/BiRefNet-general-bb_swin_v1_tiny-epoch_232.onnx", 84 | "format": "onnx", 85 | "location": "hivision/creator/weights", 86 | }, 87 | "retinaface-resnet50": { 88 | "url": "https://github.com/Zeyi-Lin/HivisionIDPhotos/releases/download/pretrained-model/retinaface-resnet50.onnx", 89 | "format": "onnx", 90 | "location": "hivision/creator/retinaface/weights", 91 | }, 92 | } 93 | 94 | # 如果选择下载所有模型 95 | if "all" in models_to_download: 96 | selected_urls = model_urls 97 | else: 98 | selected_urls = {model: model_urls[model] for model in models_to_download} 99 | 100 | if not selected_urls: 101 | print("No valid models selected for download.") 102 | return 103 | 104 | download_models(selected_urls) 105 | 106 | 107 | if __name__ == "__main__": 108 | MODEL_CHOICES = [ 109 | "hivision_modnet", 110 | "modnet_photographic_portrait_matting", 111 | # "mnn_hivision_modnet", 112 | "rmbg-1.4", 113 | "birefnet-lite", 114 | "all", 115 | ] 116 | 117 | parser = argparse.ArgumentParser(description="Download matting models.") 118 | parser.add_argument( 119 | "--models", 120 | nargs="+", 121 | required=True, 122 | choices=MODEL_CHOICES, 123 | help='Specify which models to download (options: hivision_modnet, modnet_photographic_portrait_matting, mnn_hivision_modnet, rmbg-1.4, all). Only "all" will download all models.', 124 | ) 125 | args = parser.parse_args() 126 | 127 | models_to_download = args.models if args.models else ["all"] 128 | main(models_to_download) 129 | -------------------------------------------------------------------------------- /test/create_id_photo.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding: utf-8 -*- 3 | r""" 4 | @DATE: 2024/9/5 21:39 5 | @File: create_id_photo.py 6 | @IDE: pycharm 7 | @Description: 8 | 用于测试创建证件照 9 | """ 10 | from hivision.creator import IDCreator 11 | import cv2 12 | import os 13 | 14 | now_dir = os.path.dirname(__file__) 15 | image_path = os.path.join(os.path.dirname(now_dir), "app", "images", "test.jpg") 16 | output_dir = os.path.join(now_dir, "temp") 17 | 18 | image = cv2.imread(image_path) 19 | creator = IDCreator() 20 | result = creator(image) 21 | cv2.imwrite(os.path.join(output_dir, "result.png"), result.standard) 22 | cv2.imwrite(os.path.join(output_dir, "result_hd.png"), result.hd) 23 | -------------------------------------------------------------------------------- /test/temp/.gitkeep: -------------------------------------------------------------------------------- 1 | 存放一些测试临时文件 --------------------------------------------------------------------------------