criteria 0.1.4

Deep learning loss functions and models for image similarity

Image Similarity Criteria

A Python package that provides various perceptual similarity metrics for comparing images using state-of-the-art models including Face Recognition Systems (FRS), CLIP, and LPIPS.

Features

• Face Recognition System (FRS) based identity loss with multiple model options:
  • IR-152
  • IR-SE-50
  • MobileFaceNet
  • FaceNet
  • CurricularFace
• CLIP-based similarity metrics
• LPIPS (Learned Perceptual Image Patch Similarity)
• Ensemble ID loss for combining multiple FRS models

Installation

pip install criteria

Usage

Face Recognition System (FRS)

from criteria import id_loss
from PIL import Image

# Initialize ID loss with a specific model
id_loss_fn = id_loss.IDLoss(model_name='ir_se50')

# Load images
img1 = Image.open('image1.jpg')
img2 = Image.open('image2.jpg')

# Calculate identity loss
loss = id_loss_fn(img1, img2)

CLIP Similarity

from criteria import clip_loss

# Initialize CLIP loss
clip_loss_fn = clip_loss.CLIPLoss()

# Calculate CLIP similarity
similarity = clip_loss_fn(img1, img2)

LPIPS

from criteria import lpips_loss

# Initialize LPIPS
lpips_fn = lpips_loss.LPIPSLoss()

# Calculate perceptual similarity
distance = lpips_fn(img1, img2)

Model Weights

Pre-trained model weights will be automatically downloaded when initializing the respective loss functions.

License

MIT License

Citation

If you use this package in your research, please cite:

@misc{criteria2023,
  author = {Le, Minh-Ha},
  title = {Criteria: Image Similarity Metrics},
  year = {2023},
  publisher = {GitHub},
  url = {https://github.com/minha12/criteria}
}