[CS420] Machine Learning: Free-hand Sketch Classification Problem

Introduction

This is SJTU machine learning course group project code. The content of this project is free-hand sketch classification problem. The data we used in this project is a part of data from The Quick, Draw! Dataset.

Datatools

datatools folder includes vec2pix.py and padding.py these two files.
vec2pix.py transforms the sequentially recorded data in the source dataset into vector images in SVG format, and then into PNG images. Finally, gray sampling is carried out to obtain a 28*28 matrix to represent an image.
padding.py fills in the 28*28 image. It fills blank pixels around it to make it a 32*32 image, which is easy to use as input for the pre-training model.

Models

models folder includes classification_CNN.py and classification_pretrained_model.py these two files.
classification_CNN.py is the CNN model built by us manually, including three convolutional layers, pooling layers and two fully connected layers. In classification_pretrained_model.py, we used two pre-trained models in tensorflow.keras.application: MobileNet_v2 and ResNet50.