• Setup
• Car Classifier
• HomeWork

• Currently using submodule keras-YOLOv3-model-set

  # Clone this repository
  git clone git://github.com/CSenshi/MV.git
  
  # Change directory to repository
  cd MV
  
  # Download keras submodule
  # destination directory: $(pwd)/YOLO
  git submodule update --init --recursive

• Create venv

  python3 -m venv ./venv

• Activate venv

  source venv/bin/activate

• Install

  sudo apt install python3-opencv
  pip install Cython
  pip install -r requirements.txt

• Deactivate venv

  deactivate

• Download Weights

  # change direcotry to keras-yolo submodule
  cd YOLO
  
  # Download Weights
  wget -O weights/yolov3-tiny.weights https://pjreddie.com/media/files/yolov3-tiny.weights
  
  # Convert to teras weights
  python  tools/model_converter/convert.py \
          cfg/yolov3-tiny.cfg \
          weights/yolov3-tiny.weights \
          weights/yolov3-tiny.h5

• Run Jupyter Notebook

  jupyter notebook

• It seems notebook rendering is problem in github. Try reloading few times, if problem still occures please open it via following link:
  https://nbviewer.jupyter.org/

script detect_car_color.ipynb is responsible for clasifying each car's pictures as interior or exterior.

• Model used: YOLO_np
• Wrapper function used: predict()

Before running script:

  ...
  ├── 45976134
  ├── 45976135
  │   ├── 1.jpg
  │   ├── 2.jpg
  │   ├── 3.jpg
  │   ├── 4.jpg
  │   ├── 5.jpg
  │   ├── 6.jpg
  │   └── 7.jpg
  ├── 45976136
  ...

After script execution:

...
├── 45976134
├── 45976135
│   ├── exterior
│   │   ├── 2.jpg
│   │   ├── 5.jpg
│   │   ├── 6.jpg
│   │   └── 7.jpg
│   └── interior
│       ├── 1.jpg
│       ├── 3.jpg
│       └── 4.jpg
├── 45976136
...

As for homework I took 3 exercises to work on

1. Doors
2. Car Category
3. Car Color

For all of those exerises I wrote one generic script that would work on all the given problems, with slight changes. As for model I created new model class that is subclass of nn.Module and as for base model I used resnet18.

• Why Resnet?

  As for base model I had to choose between many models, but finally decided to choose between 3 models: Resnet, VGG, Densnet. I tried all of those model and results showed that VGG performs poorly (expected better results), so i had to throw it away. as for Densnet it is great model, but it was taking a lot of time, but model result wasn't as different as elapsed time for each model. So I choose to stay with model that takes small time (RESNET).

• Why 18 layers?

  Resnet comes with 18, 34, 50, 101 and 152 layers. As results showed prediction is getting better only after 50 layers, but as it was in modeling, there is downside of time. It takes a lot of time than resnet 18, but for 34 layer it didn't act as good as i expeted. So after many tests I decided to stick with Resnet model with 18 layers, which is enough for given problems

As for loss function I used torch.nn.CrossEntropyLoss(), This criterion combines nn.LogSoftmax() and nn.NLLLoss() in one single class (softmax is a MUST for this problem because we are having problem with multiple classes).

• Gradient

  for gradient descent I used Stochastic Gradient Descent (SGD) from torch.optim. so if you use a typical Gradient Descent optimization technique on big data(samples), you will have to use all of the samples for completing one iteration while performing the Gradient Descent, and it has to be done for every iteration until the minima is reached. Hence, it becomes computationally very expensive to perform. In SGD, it uses only a single sample, i.e., a batch size of one, to perform each iteration. The sample is randomly shuffled and selected for performing the iteration.

• Scheduler

  torch.optim.lr_scheduler provides several methods to adjust the learning rate based on the number of epochs.