JunkWaxHero 🦸‍♂️

JunkWaxHero is your go-to computer vision model for identifying "Junk Wax" baseball card sets from the golden era of 1980 to 1999. Whether you're a collector, dealer, or enthusiast, JunkWaxHero simplifies the process of cataloging and verifying your collection with high precision and recall. The purpose of JunkWaxHero is to provide a model that quickly identifies baseball card sets without relying on SaaS-based inference APIs.

📦 Available Formats

JunkWaxHero is available in ONNX format to cater to different deployment needs and preferences.

ONNX: Perfect for interoperability across various platforms and optimizing for different hardware accelerators.

📊 Training Details

Model Version: Iteration 22
Domain: General (compact) [S1]

Overall Performance Metrics

Precision: 99.0%
Recall: 96.7%
mAP: 98.5%

📁 Dataset

The training dataset for JunkWaxHero was meticulously curated to ensure robustness and versatility. It includes baseball card images captured on multiple different surfaces and backgrounds, with varying zoom levels to simulate real-world conditions. All images adhere to a 3:4 aspect ratio to align with modern cell phone portrait modes. The baseball cards in the dataset are vertically aligned to maintain consistency during training. Support for horizontally aligned cards photographed in their horizontal orientation is planned for future iterations to enhance the model's flexibility.

🎯 Performance Per Set

JunkWaxHero has been meticulously trained to recognize the following baseball card sets with exceptional accuracy:

Tag	Precision	Recall	Average Precision (AP)
1982 Donruss	100.0%	100.0%	100.0%
1984 Topps	100.0%	100.0%	100.0%
1987 Fleer	100.0%	95.5%	95.5%
1987 Topps	100.0%	100.0%	100.0%
1988 Donruss	95.5%	100.0%	100.0%
1988 Fleer	100.0%	71.4%	90.5%
1988 Fleer Pack	100.0%	100.0%	100.0%
1988 Score	100.0%	100.0%	100.0%
1988 Topps	100.0%	95.8%	100.0%
1989 Bowman	100.0%	94.7%	99.7%
1989 Donruss	100.0%	94.7%	100.0%
1989 Fleer	100.0%	100.0%	100.0%
1989 Score	100.0%	100.0%	100.0%
1989 Topps	95.0%	90.5%	95.0%
1989 Upper Deck	100.0%	92.3%	100.0%
1990 Donruss	100.0%	100.0%	100.0%
1990 Fleer Pack	92.3%	100.0%	100.0%
1990 Leaf	94.7%	94.7%	99.3%
1990 Leaf Pack	100.0%	100.0%	100.0%
1990 Topps	100.0%	100.0%	100.0%
1990 Upper Deck High Series Pack	100.0%	95.7%	99.5%
1991 Fleer Ultra	100.0%	100.0%	100.0%
1991 Leaf	100.0%	90.9%	95.5%
1991 Leaf Pack	100.0%	100.0%	100.0%
1991 Upper Deck	100.0%	89.5%	94.7%
1991 Upper Deck Low Series Pack	100.0%	96.0%	99.6%
1992 Fleer	100.0%	100.0%	100.0%
1992 Fleer Ultra	100.0%	100.0%	100.0%
1992 Fleer Pack	100.0%	75.0%	75.0%
1992 O-Pee-Chee Premiere	100.0%	94.7%	94.7%
1992 Pinnacle	100.0%	100.0%	100.0%
1992 Pinnacle Pack	100.0%	85.7%	100.0%
1992 Upper Deck	100.0%	87.5%	95.7%
1992 Upper Deck High Series Pack	100.0%	100.0%	100.0%
1993 Fleer Series 1 Pack	100.0%	100.0%	100.0%
1993 Fleer Series 2 Pack	100.0%	100.0%	100.0%
1993 Topps	95.2%	90.9%	99.0%
1994 Leaf	100.0%	100.0%	100.0%
1994 Pinnacle	100.0%	100.0%	100.0%
1994 Score	100.0%	100.0%	100.0%
1995 Leaf	100.0%	100.0%	100.0%
1995 Select	95.5%	100.0%	100.0%
1996 Pinnacle	92.3%	100.0%	100.0%

Included sets are marked with a ✅, while excluded sets (if any) would be marked with a ❌.

🚀 Getting Started

Repository

Access the repository on Hugging Face: https://huggingface.co/enusbaum/JunkWaxHero/

Installation

To integrate JunkWaxHero into your projects, choose the format that best fits your workflow:

ONNX

Install ONNX Runtime:
```
pip install onnxruntime
```
Download the ONNX Model:

Navigate to the ONNX directory in the repository and download the model.onnx file from https://huggingface.co/enusbaum/JunkWaxHero/tree/main/onnx.

Load the ONNX Model:

import onnxruntime as ort
import numpy as np

# Path to the downloaded ONNX model
MODEL_PATH = 'path/to/junkwaxhero/onnx/junkwaxhero.onnx'

# Initialize the ONNX runtime session
session = ort.InferenceSession(MODEL_PATH)

# Get input and output names
input_name = session.get_inputs()[0].name
output_name = session.get_outputs()[0].name

# Example function to run inference
def predict(image):
    image = np.expand_dims(image, axis=0).astype(np.float32)  # Add batch dimension and ensure correct type
    predictions = session.run([output_name], {input_name: image})
    return predictions

Usage

Prepare Your Images: Ensure your baseball card images are clear and well-lit for optimal identification. Images should be in a 4:3 ratio and vertically aligned.
Load the Model: Use the appropriate code snippet above based on your chosen format (TensorFlow or ONNX).

Identify Sets: Pass your images through JunkWaxHero to accurately identify the set they belong to.

from PIL import Image

# Load and preprocess the image
def load_image(image_path):
    image = Image.open(image_path)
    image = image.resize((width, height))  # Resize to the input size expected by the model
    image = np.array(image)
    return image

# Example usage
image = load_image('path/to/baseball_card.jpg')
predictions = predict(image)

# Get the predicted label
predicted_label = labels[np.argmax(predictions)]
print(f'Predicted Set: {predicted_label}')

Note: Replace (width, height) with the actual dimensions expected by your model.

Optimization

Leverage the high precision and recall rates to maintain an organized and verified collection. Depending on your deployment environment, choose between TensorFlow for seamless integration with TensorFlow ecosystems or ONNX for broader platform compatibility and performance optimizations.

🔍 Use Cases

Collectors: Quickly verify and catalog your baseball card collection.
Dealers: Streamline inventory management with accurate set identification.
Enthusiasts: Explore and discover sets with ease using advanced computer vision techniques.

🛠️ Technical Details

JunkWaxHero leverages state-of-the-art computer vision architectures optimized for identifying baseball card sets from images. Trained on a diverse dataset spanning multiple decades, the model ensures robustness and reliability across various card conditions and designs. By offering both TensorFlow and ONNX formats, JunkWaxHero provides flexibility for different deployment scenarios, whether you're integrating into existing TensorFlow pipelines or seeking cross-platform compatibility with ONNX.

📄 License

This project is licensed under the MIT License.

📫 Contact

For questions, suggestions, or support, please reach out to [email protected].

Happy Collecting!

enusbaum
/

JunkWaxHero