iMarket_README

iMarket is a Fruit freshness evaluation tools based on YOLOv5 and Pretrained vision network. It’s forked from Ultralytics’ YOLOv5 repository and based on this, added and customized for our purpose.

How to use

---

• Install

  Clone repo and install requirements.txt in a Python≥3.7.0 environment, including PyTorch≥1.7.

  git clone https://github.com/seki5405/iMarket_Fruit_Spoilage_Detection.git
  cd iMarket_Fruit_Spoilage_Detection
  pip install -r requirements.txt

• Training

  • Train regression model

    • freshness_train.py is for training your regression model
    • You have to train for each fruit with its own classified dataset
    • There are commented out codes for classification approach(#classfication)

    python3 freshness_train.py --save-name 'pretrained weights path'\
    													  --epochs 50 --dataset 'dataset path'
    /* Parser information
    '--base-model', type=str, default='vgg16', help='Base model for the regression model'
    '--epochs', type=int, default=50, help='Training epochs'
    '--batch-size', type=int, default=32, help='Training batch size'
    '--optimizer', type=str, choices=['SGD', 'Adam', 'AdamW'], default='Adam', help='optimizer'
    '--save-name', type=str, required=True, help='Name to save weights after training'
    '--dataset', type=str, required=True, help='Dataset path'
    '--imgsz', '--img', '--img-size', type=int, default=416, help='Image size (width = height)'
    '--split', type=float, default=0.2, help='train_valid split ratio' */

  • Train YOLOv5 based model

    • yolo_train.py is for training Object Detection model
    • It’s mainly inherited from YOLOv5 except some customized functions

    !python yolo_train.py --img 416 --epochs 400 --batch 32 \
    										  --data 'yaml_path' --cfg models/yolov5s.yaml  \
    											--weights yolov5s.pt --name 'save name' \
    // You can change the arguments and add above this

• Prediction

  imarket_main.py is the main function to implement evaluation for each fruits

  • To visualize the results on colab, use the cod below

    // For visualizaiton in colab, use this code
    import cv2
    from google.colab.patches import cv2_imshow
    
    def show_img(url):
      img_name = url.split('/')[-1]
      img_path = os.path.join("saved dir")
      for path, dir, fname in os.walk('saved dir'):
        if img_name in fname:
          f_path = os.path.join(path, img_name)
      img = cv2.imread(f_path)
      cv2_imshow(img)

  • To implement the main function

    URL = "Your own image url"
    python imarket_main.py --weights $yolo_path --freshness-weights $reg_path --imgsz 416 --conf 0.25 --source $URL
    show_img(URL)

  • Example of result