seedmanc's picture
Update README.md
1c5c466
metadata
title: Batch Laion Aesthetic Predictor
emoji: 👁️⃤
colorFrom: blue
colorTo: red
sdk: gradio
sdk_version: 3.18.0
app_file: opp.py
pinned: true
fullWidth: true
duplicated_from: Geonmo/laion-aesthetic-predictor

Computes several aesthetic scores statistics over a bunch of pictures (min, max, range, average, median, coeff. of variance), exporting the results in CSV.

Currently uses the sac+logos+ava1-l14-linearMSE.pth MLP trained on 224x224 images from CLIP checkpoint ViT-L/14. Should work on CPU and CUDA, though I can only test for CPU here.

Colab for more configurability (ViT-L/14@336px available) and free GPU acceleration.