johndpope / MTG_CLIP

MTG IMAGE RETRIVAL

WORK IN PROGRESS This repo uses CLIP in order to detect similar images.

How to install:

1. Run 'get_dataset.py' with yout favourite editions and get images
2. Run 'trainer.py' to train the model
3. Run 'predictor.py' with you image.

Files and folders

• dataset: where the images are stored (output get_dataset.py)
• embeddings: where the images embeddings are stored (output trainer.py)
• images: where input images and results are stored (output predictor.py)
• pca.npz: binary file that describes the PCA used to reduce feature in model, needed for inference (output of imgebeddings model)

Results:

IMPORTANT:

This repo needs Tensorflow, Pytorch or Flax installed.

TODO:

• docker
• API
• Benchmark
• fix predictor output