Awesome

Match Steam Banners

This repository contains Python code to retrieve Steam games with similar store banners.

It is based on the best practices observed in my previous repository download-steam-banners:

• features extracted by a neural network (MobileNet v1 previously ; v3 here),
• global average pooling of features, as in MobileNet papers and implementations,
• cosine similarity.

Requirements

• Install the latest version of Python 3.X.
• Install the required packages:

python -m pip install --upgrade pip
pip install --upgrade cython
pip install -r requirements.txt -f https://download.pytorch.org/whl/torch_stable.html

NB: Tensorflow 2 supports Python 3.5–3.8, not 3.9.

• Install OpenAI's CLIP with pip:

pip install git+https://github.com/openai/CLIP.git

• Install Facebook's DINO with git:

git clone https://github.com/facebookresearch/dino.git
mv dino/vision_transformer.py .
mv dino/utils.py .

Model

MobileNet is a convolutional neural network, trained for classification on ImageNet-1k (1.28M images with 1000 classes).

In this repository, the image encoder is MobileNetV3-Small.

Data

Data is available in download-steam-banners-data/.

The most recent data snapshot was downloaded in August 2020 with this Colab notebook. It consists of 19,049 vertical Steam banners resized from 300x450 to 256x256 resolution.

Usage

NB: a Colab notebook is available in my colab branch.

1. Features

First, compute and store the 1024 features corresponding to each banner:

python build_feature_index.py

2. Similar games

Find the 10 most similar store banners to curated query appIDs:

python retrieve_similar_features.py

NB: by default, query appIDs consist of:

• the top 100 most played games during the past two weeks, according to SteamSpy,
• a few manually curated games.

3. Unique games

Find the one most similar store banner to all appIDs available on the store, then display the most unique games:

python find_unique_games.py

NB: unique games are ones which are the most dissimilar (low similarity score) to others to their first neighbor.

4. Export data and results for a web app

Optionally, export data and results for a web app:

python export_data_for_web_app.py

Results

Results obtained with MobileNet v3 are shown on the Wiki.

The linked pages contain a lot of images and might be slow to load depending on your Internet bandwidth.

Similar games

Direct links to similarity results are available below:

• for each game, find the 10 most similar games.

For instance:

Or:

Unique games

Direct links to similarity results are available below:

• for each unique game, display the 1 most similar game,
• a grid of unique games.

Unique games seem to have in common that their banner contains mostly a texture and a title, without any large character, item or shape.

For instance:

References

• download-steam-banners: retrieve Steam games with similar store banners,
• download-steam-screenshots: retrieve Steam games with similar store screenshots,