Simple Clothing Classifier

Pure TorchVision-based implementation of clothing classification.

Features

• 64px per 64px input images supported
• Train a model on Fashion MNIST dataset
• Multithreading support (may handle multiple requests at the same time)
• 10 basic classes (limited due to Fashion MNIST)

Requirements

Before you begin, ensure you have a machine with an GPU that supports modern CUDA, due to the computational demands of the docker image.

• Nvidia GPU
• CUDA
• Docker
• Docker Compose
• Nvidia Docker Runtime

For detailed instructions on how to prepare a Linux machine for running neural networks, including the installation of CUDA, Docker, and Nvidia Docker Runtime, please refer to the publication “How to Prepare Linux for Running and Training Neural Networks? (+ Docker)” on Russian.

Installation

1. Clone the repo and switch to sources root:

   git clone https://github.com/EvilFreelancer/cloth-classifier.git
   cd cloth-classifier
   

2. Copy the provided Docker Compose template:

   cp docker-compose.dist.yml docker-compose.yml
   

3. Build the Docker image:

   docker-compose build
   

4. Start the services:

   docker-compose up -d
   

How to use

curl \
  "http://localhost:8080/predict" \
  -X POST \
  -H "Content-Type: multipart/form-data" \
  -F "file=@tshirt.jpg"

{
  "class": 0,
  "class_label": "T-shirt",
  "probability": 2.12345
}

Available classes

CLASSES = [
    'T-shirt', 'Trouser', 'Pullover', 'Dress', 'Coat',
    'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle Boot'
]

Train model

Login to container:

docker-compose exec app bash

Run training script:

python3 /app/train_model.py

It will train a model and put it to ./models/cloth_model_simple.pth.

Links

• https://github.com/sssingh/fashion-mnist-classification/blob/master/fashion_mnist_classification_nn_pytorch.ipynb
• https://github.com/roboflow/notebooks/blob/main/notebooks/train-vision-transformer-classification-on-custom-data.ipynb