Intro[Custom object detection](Using YOLOv2)

To get started with working on OTS-retail project git clone or download the darflow repository first. darkflow

---

Assuming that the installation steps from the darflow repository has been followed we now outline the various steps involved in this project.

---

• First step is to create a dataset--can use tools like labelImg,labelme...etc.I prefer labelImg! (Skip this if you already have a dataset and their annotations)call

• Second step is :

  1. Case 1(Training on a GPU server):

  • If training on a GPU server then first set up the environment.

  • Convert the annotations format to the GPU based format by using the script from my repository- xml_preprocess_path.py

  2. Case 2(Training on our local computer):
  • Convert the annotations format to the local computer based format by using the script from my repository- xml_preprocess_path.py

• Training on our own dataset

  • Create a copy of the configuration file tiny-yolo-voc.cfg and rename it according to your preference.In my case there is only one class so tiny-yolo-voc-1c.cfg (It is crucial that you leave the original tiny-yolo- voc.cfg file unchanged.)

  • In tiny-yolo-voc-1c.cfg, change classes in the [region] layer (the last layer) to the number of classes you are going to train for (Line 120: classes=no of classes)

     ...     
     [region]
     anchors = 1.08,1.19,  3.42,4.41,  6.63,11.38,  9.42,5.11,  16.62,10.52
     bias_match=1
     classes=1
     coords=4
     num=5
     softmax=1
     ...
  

  • In tiny-yolo-voc-1c.cfg, change filters in the [convolutional] layer (the second to last layer)[Line:114] to num *(classes + 5).In our case, num is 5 and classes are 3 so 5 * (1 + 5) = 30 therefore filters are set to 30.

     ...
     [convolutional]
     size=1
     stride=1
     pad=1
     filters=30
     activation=linear
  
     [region]
     anchors = 1.08,1.19,  3.42,4.41,  6.63,11.38,  9.42,5.11,  16.62,10.52
     ...
  

  • Change labels.txt to include the label(s) you want to train on (number of labels should be the same as the number of classes you set in tiny-yolo-voc-3c.cfg file). In our case, labels.txt will contain 3 labels.

    label1
    label2
    .
    .
    .
  

  • Third step is(execute in the terminal the following command to start training):

    python flow --model cfg/tiny-yolo-voc-1c.cfg --load bin/tiny-yolo-voc.weights --train --annotation /Path/to/annotations --dataset /Path/to/images --epoch 100 --summary /Path/to/store/logs/for/tensorboard/visualization.
    

  • To visualize the training in the tensorboard:

  tensorboard --logdir logstrain
  
  

• After the training has been completed you should see the trained file configurations with the name tiny-yolo-voc- 1c-14250.profile which is can now be used for object detection.

• From the repository in object_rec.py file configure the model trained and the checkpoint file.(Also configure the path of the image on which you want the object detector to detect.

  options = {
  'model': 'cfg/tiny-yolo-voc-1c.cfg',
  'load': 14250,
  'threshold': 0.25,
  'gpu': 1.0(if using a gpu)
   }
   
  

• Finally execute :

   
   python object_rec.py