Automatic Review System based on Deep Neural network

This project is to create an automatic review system over user taken images. My group is working on the simplest encoder-decoder model, where the encoder is achieved by Convolutional Neural Network (CNN) and decoder is achieved by Recurrent Neural Networks (RNN).
The project includes both computer vision (CV) and natual language processing (NLP) disciplines.

Key Concepts

• Convolutional Neural Network (CNN)
• InceptionV3 Feature
• Recurrent Neural Networks (RNN)
• Long short-term memory (LSTM)
• GloVe Embeddings

Key Technology Takeaways

• Python
• Pandas, Numpy
• PyTorch, Keras
• Use of GPU
• Linux System

Contribution

The project was completed by a group of 7 members, who were divided into 3 teams and took charge of training different models to compare.Team A, where I was leading 3 members, worked on an encoder-decoder architecture model without attention mechanism. I was responsible for most of data processing, programming and model tuning tasks for my team.

Individual Presentation Slides

Result Example

Detailed Report

Source Code