dudcjs2779's picture
Create README.md
500667e verified
metadata
license: mit
base_model:
  - timm/eva02_base_patch16_clip_224.merged2b_s8b_b131k
pipeline_tag: zero-shot-image-classification
library_name: open_clip

anime-style-tag-clip

This model weight is based on eva02_base_patch16_clip.
The dataset size is small, which may result in slightly reduced generalization performance of the model.

Dataset

I trained this model with anime-style image-tag data.
Train: 29187
Valid: 3000

Validation results

R@1: 0.87733
R@5: 0.96767
R@10: 0.98233

Training & Inference code

https://github.com/dudcjs2779/tag-clip.git