/PSA-Baseball-Grades

A dataset of baseball cards with their associated Professional Sports Authenticator (PSA) grades.

Primary LanguagePython

PSA-Grades-Baseball

PSA-Grades-Baseball

PSA-Grades-Baseball is a dataset of baseball card images and associated Professional Sports Authenticator (PSA) grades. PSA-Grades-Baseball contains 1150 baseball card images for each of the 10 major PSA grades totaling 11500 images.

Data Collected With

Getting Started

These instructions will get you a copy of the dataset on your local machine.

Installation

The following are different options to download the dataset:

GitHub

  1. Clone the repository.
git clone https://github.com/samsilverman/PSA-Grades-Baseball.git
  1. Optional: If using automatic dataset generators such as tf.keras.preprocessing.image_dataset_from_directory, you may want to copy the psa directories from the repository to a separate directory to avoid miscellaneous files (i.e. git files) from being mistaken for classes.

Zip

  1. Download the latest version of the dataset from Releases.

Details

The following steps were applied to each scrapped image to ensure a consistent format for all images in the dataset:

  1. Ensure a consistent image size of 150x200. This was done in a three-step process (code in dataset_processor.py):

    1. Images with width > height were rotated clockwise by 90°.

    2. Images were scaled to 150x200 (or as possible).

    3. Images that could not be exactly scaled to 150x200 were padded using replication of pixel values.

  2. A simple file structure was implemented for ease of use:

📦 PSA-Grades-Baseball/
    📁 psa1/
        🖼 psa1_1.jpg
        🖼 psa1_2.jpg
        ...
        🖼 psa1_1150.jpg
    📁 psa2/
    ...
    📁 psa10/

Known Issues

The following are a list of known issues with the dataset that should be addressed:

  1. The dataset may contain "bad" data which consists of one or more of the following:

    1. Duplicate images

    2. Incorrect image grades

    3. irrelevant images

  2. The dataset contains unimportant features which consist of one or more of the following:

    1. Hands and tables

    2. Colorful and detailed backgrounds

    3. PSA grade cases and labels

AutoGrade: A Deep Learning Approach to Trading Card Grading

AutoGrade is a deep learning system that was trained on PSA-Grades-Baseball to provide estimates on PSA grades. AutoGrade and PSA-Grades-Baseball were developed as a final project for the course COMS 4995 (see Acknowledgements for more).

View AutoGrade Video 🎥

Contact

Sam Silverman - @sam_silverman - samuel.silverman@columbia.edu

Project Link: https://github.com/samsilverman/PSA-Grades-Baseball

Acknowledgements

PSA-Grades-Baseball was created for a final project for the course COMS 4995: Deep Learning for Computer Vision taught by Prof. Peter Belhumeur.