A real-time object detector API for floating plastic litter, built with YOLOv4 and Keras/Tensorflow.

FLOTVIS .  
├── VOCdevkit               # Main training dataset  
│   └── VOC2007  
│       ├── Annotations  
│       ├── JPEGImage  
│       └── ImageSet  
│           └── Main  
├── model_data              # weights and anchors  
│   ├── yolov4_anchors.txt  
│   ├── yolov4_weights.h5   
│   └── ...  
├── nets                    # backbone and evaluation functions  
│   ├── CSPdarknet53.py  
│   ├── loss.py  
│   ├── ious.py  
│   └── yolo_net.py  
├── test                    # dataset for testing  
│   ├── test1.jpg  
│   └── ...  
└── utils                   # utilities  
    ├── util.py  
    └── gputil.py  

tensorflow==1.15
Keras==2.1.5
scipy==1.2.1
numpy==1.17.0
matplotlib==3.1.2
opencv_python==4.1.2.30
tqdm==4.60.0
Pillow==8.2.0
h5py==2.10.0

# predict single image
$ python3 main.py --mode='image' --img='test/test.jpg'

# predict all images in folder
$ python3 main.py --mode='batch' --imgdir='./test/'

# predict video
$ python3 main.py --mode='video' --vid='test/test.mp4'

# predict using camera
$ python3 main.py --mode='camera'

# get FPS
$ python3 main.py --mode='fps'

• LabelImg for creating ground truth bounding box default output in .xml
  class(es): plastic
  yolo_net data format: class_number <x_center> <y_center> <width> <height>

# get FPS
$ python3 train.py

• mean Average Precision (mAP)
• Complete-IoU Loss and Cluster-NMS
• ImgAug
• LabelImg
• Semantic Versioning
• tqdm
• How to improve object-detection
• A neat implementation of YOLO v3 with TF 2.2 and Keras
• A popular Implementation of YOLO v3 based on multi-backend Keras
• Another Implementation of YOLO v3 based on multi-backend Keras
• YOLO implementation on Keras/TF2
• YOLO implementation on Keras
• YOLOv4 implementation on Keras
• YOLOv4 implementation on Keras
• YOLOv4 implementation on Keras
• Deep-SORT-YOLOv4
• Web Based YOLOv4 Graphic Interface by BMWInnovationLab

• If Jupyter Notebook running on your local machine prompts "No Module named Tensorflow": it is a problem of Jupyter with virtualenv.

  $ cd $HOME  
  $ brew install  
  $ pip3 install jupyter  
  $ jupyter notebook --generate-config  
  # this installs kernelspec python3 in $HOME/Library/Jupyter/kernels/python3  
  $ python3 -m ipykernel install --user  

  Assuming you use pyenv and works in a virtual environment named venv, then:

  # Activate your virtualenv  
  $ pyenv activate tf1.15  
  # Check path of the Python interpreter  
  $ pyenv which python  
  $HOME/.pyenv/versions/tf1.15/bin/python # copy this output  
  # Deactivate the virtualenv  
  $ pyenv deactivate  
  $ mkdir $HOME/Library/Jupyter/kernels/venv
  $ touch $HOME/Library/Jupyter/kernels/venv/kernel.json

  add the following content to the newly created kernel.json

  {
    "argv": [
      "$HOME/.pyenv/versions/tf1.15/bin/python",
      "-m", "ipykernel",
      "-f", "{connection_file}"
    ],
    "display_name": "venv",
    "language": "python"
  }

  Finally, check the jupyter kernel using:

  jupyter kernelspec list