jason-li-831202 / object-segmentation-web

This project used Yolov8/AnimeGAN and Flask to accomplish the task of background segmentation , background remove and background replacement. Additionally, you can change different display styles.

Object Background Removal Web using YoloV8 with Flask

Web application for real-time object segmentation using Flask , YOLOv8, AnimeGANv2/v3 model in ONNX weights.

After obtaining images through the camera, it is possible to separate the target and background in the scene, and composite the target with a Web URL/Local Path video background. Additionally, you can change different display styles and save a screenshot.

➤ Contents

1. Requirements

2. ONNX-Model

3. Examples

4. Docker

5. Demo

6. Updates

7. License

➤ Requirements

• OpenCV, Flask, gevent, onnxruntime and youtube-dl.

• Install :

  The requirements.txt file should list all Python libraries that your notebooks depend on, and they will be installed using:

  pip install -r requirements.txt
  

• Note :

  If you use a YouTube URL as the link to replace your background, please make sure to modify the following.

  1. YouTube Unable to extract uploader id, so you need to revise [Your Path]\site-packages\youtube_dl\extractor\youtube.py :

  • 'uploader_id': self._search_regex(r'/(?:channel|user)/([^/?&#]+)', owner_profile_url, 'uploader id') if owner_profile_url else None,

    $\Downarrow$

  • 'uploader_id': self._search_regex(r'/(?:channel/|user/|@)([^/?&#]+)', owner_profile_url, 'uploader id', default=None),

  2. Youtube does no longer have a like/dislike count, so you need to comment out [Your Path]\site-packages\pafy\backend_youtube_dl.py :

  • self._likes = self._ydl_info['like_count']

  • self._dislikes = self._ydl_info['dislike_count']

➤ ONNX-model

You can convert the YOLOv8-seg model to ONNX using the following Google Colab notebook :

• The License of the models is GPL-3.0 license: License

• Quantize ONNX models :

  Converting a model to use float16 instead of float32 can decrease the model size.

  python onnxQuantization.py -i <path-of-your-onnx-model>
  

➤ Examples

• Setting Config :

  model_config = {
      "model_path": 'models/yolov8n-seg-coco.onnx', # model path
      "classes_path" : 'models/coco_label.txt', # classes path
      "box_score" : 0.4,
      "box_nms_iou" : 0.45,
      "box_aspect_ratio" : None,
      "box_stretch" : None,
  }
  
  cam_config = {
      "cam_id" : 0,
      'exposure': -2, # init cam exposure
      'contrast': 50 # init cam contrast
  }

  After running, the config information will appear above the menu :

  

• Run :

  python Application.py
  ... 
  The server will be accessible at [ http://localhost:8080 ].
  

  Display Mode	Describe
  DisplayType.NONE	Just show your webcam image.
  DisplayType.BASIC_MODE	Show detect and segmentation target results on image.
  DisplayType.DETECT_MODE	Separate the target box and background on image.
  DisplayType.SEMANTIC_MODE	Separate the target segmentation and background on image.

➤ Docker

From the root directory, run :

docker build -t <name-of-your-image> -f ./docker/Dockerfile .
docker run --rm --privileged --device="/dev/video0:/dev/video0" -t -p 8080:8080 <name-of-your-image>

➤ Demo

• Demo Youtube Video

• Display Mode Switch

  

• Display Style Switch

  

• Display Background Removal

  

➤ Updates

• 2023/05/05 - Added images with downloadable transparent backgrounds.
• 2023/11/22 - Fixed youtube_dl bug, and remove pafy package.
• 2023/11/26 - Fixed custom diff model size.

➤ License

WiFi Analyzer is licensed under the GNU General Public License v3.0 (GPLv3).

GPLv3 License key requirements :

• Disclose Source
• License and Copyright Notice
• Same License
• State Changes

object-segmentation-web's Issues

The approach to great projects

I'm just an AI student, but I'm really impressed with the projects from ADAS to here. And the Chinese are really good. Can you suggest me how to approach repos like this regarding CV. I don't know where to start, I just want to improve my ability in Computer Vision. 非常感谢