anaaa2/visual-moral-compass

Visual Moral Compass

Visual Moral Compass is a fine-tuned CLIP model that classifies images based on Moral Foundations Theory.

Model Description

This model extends CLIP (openai/clip-vit-base-patch16) with five classifier heads to predict moral dimensions in images:

• Care vs. Harm: Concerns about suffering and protection
• Fairness vs. Cheating: Concerns about justice and reciprocity
• Loyalty vs. Betrayal: Concerns about group membership and solidarity
• Respect vs. Subversion: Concerns about hierarchy and authority
• Sanctity vs. Degradation: Concerns about purity and contamination

Usage

from visual_moral_compass import VisualMoralCompass

# Load model
model = VisualMoralCompass.from_pretrained("YOUR_USERNAME/visual-moral-compass")

# Classify an image
results = model.classify_image("path/to/image.jpg")
print(results)

Citation

If you use this model, please cite:

@inproceedings{moralclip2025,
      author = {Condez, Ana Carolina and Tavares, Diogo and Magalh\~{a}es, Jo\~{a}o},
      title = {MoralCLIP: Contrastive Alignment of Vision-and-Language Representations with Moral Foundations Theory},
      year = {2025},
      isbn = {9798400720352},
      publisher = {Association for Computing Machinery},
      address = {New York, NY, USA},
      doi = {10.1145/3746027.3758166},
      booktitle = {Proceedings of the 33rd ACM International Conference on Multimedia},
      pages = {12399–12408},
      numpages = {10},
      location = {Dublin, Ireland},
      series = {MM '25}
    }

Model Details

• Base Model: openai/clip-vit-base-patch16
• Training Data: Social-Moral Image Database