metadata

license: mit
language:
  - en
base_model:
  - openai/clip-vit-base-patch32
pipeline_tag: image-classification
tags:
  - aesthetics
  - creativity

Aesthetic Scorer

This model predicts 7 different aesthetic metrics for images:

Overall aesthetic score
Technical quality score
Composition score
Lighting score
Color harmony score
Depth of field score
Content score

Model Details

Based on CLIP ViT-B/32 visual encoder
Fine-tuned on the PARA dataset
Returns scores between 0-5 for each aesthetic dimension

Usage

from transformers import CLIPProcessor
from aesthetic_scorer import AestheticScorer
import torch
from PIL import Image

# Load the model
processor = CLIPProcessor.from_pretrained("rsinema/aesthetic-scorer")
model = torch.load("rsinema/aesthetic-scorer/model.pt")

# Process an image
image = Image.open("your_image.jpg")
inputs = processor(images=image, return_tensors="pt")["pixel_values"]

# Get scores
with torch.no_grad():
    scores = model(inputs)

# Print results
aesthetic_categories = ["Overall", "Quality", "Composition", "Lighting", "Color", "Depth of Field", "Content"]
for category, score in zip(aesthetic_categories, scores):
    print(f"{category}: {score.item():.2f}/5")