YouTube Title Generator

Generate titles for YouTube thumbnails using an image captioning LSTM (Long-Short Term Memory) model.

Do you watch a lot of YouTube like me and realized how thumbnails typically follow a particular style? Do you notice just how non-sensical (and sometimes clickbaity) some titles can be? Well I have and thought it would be funny to generate my own. Using this code, you can generate some quality h o t t a k e s such as these:

Ahh yes, the senate challenge with somebody from the litfam...

Description

Below are descriptions of the core files in this repo:

.
├── app
│   └── src
│       ├── generate_captions.py
│       ├── train.py
│       ├── model.py
│       ├── results.py
│       └── training_data.py

• app/src/generate_captions.py: Script to generate YouTube titles
• app/src/train.py: Caption model training script
• app/src/model.py: Caption model architecture
• app/src/results.py: View results in UI
• app/src/training_data.py: Download and prepare data

Requirements

• Python 3
• pip 3
• TensorFlow configured to use a GPU (recommended)

Getting Started

The dataset used to train this model can be found on Kaggle. If you don't have a Kaggle account, you will need to sign up for one to download the dataset. You will need USVideos.csv, however you may use the other datasets if you wish. Add the file to the app/resources/ directory.

Next, install the dependencies using:

$ make deps

Downloading and Preparing Data

To download and prepare the data used for training and testing, run the following commands:

• $ make thumbnails: Download the YouTube thumbnails
• $ make titles: Prepare the titles
• $ make features: Download the VGG16 image features

Training the Model

Train the captioning model by running the following command:

$ python3 app/src/train.py

Training the model can take a very long time. If you use a GPU, it will take considerably less time to train.

Generate Some Captions!

To see what kind of titles your model can generate, run:

$ python3 app/src/generate_captions.py

then afterwards:

$ python3 app/src/results.py

This will create a file called results.html which you can open in your browser and view your creations!

Acknowledgements

The caption model used was adapted from this tutorial over at Machine Learning Mastery.

Further reading:

• Where to put the Image in an Image Caption Generator, Tanti et al.
• What is the Role of Recurrent Neural Networks (RNNs) in an Image Caption Generator?, Tanti et al.