fashion_cnn

An online convolutional neural network (CNN) model for classifying fashion items.

The model is trained on the MNIST fashion dataset and implemented using the TensorFlow.js library.

The model is loaded on the client side and all calculations are made on the device.

Setup

npm run install
npm run dev #localhost:5173

Model caracteristics

• dataset: MNIST (fashion) (10 000 samples)
• input layer: 784 (28x28 pixels)
• convolutional layers (4):
  • 16(28x28)[5x5 filter] + stride[2] & max pool[2x2]
  • 32(14x14)[5x5 filter] + stride[2] & max pool[2x2]
• hidden layers (1): 128 neurons (ReLU activation)
• output layer: 10 (Softmax activation)
• optimizer: Adam
• loss: categoricalCrossentropy
• metrics: accuracy
• epochs: 10
• batch size: 512