app.py

import gradio as gr
import timm
import torch
import pandas as pd


TITLE = "wd-eva02-large-tagger-v3-vector"
DESCRIPTION = """
モデル：[SmilingWolf/wd-eva02-large-tagger-v3](https://huggingface.co/SmilingWolf/wd-eva02-large-tagger-v3)

日本語訳？：[p1atdev/danbooru-ja-tag-pair-20241015](https://huggingface.co/datasets/p1atdev/danbooru-ja-tag-pair-20241015)
"""

model = timm.create_model(f"hf_hub:SmilingWolf/wd-eva02-large-tagger-v3", pretrained=True)
head = model.head.weight.data
del model
df = pd.read_csv(f"https://huggingface.co/SmilingWolf/wd-eva02-large-tagger-v3/resolve/main/selected_tags.csv")
id2label = df["name"].to_dict()
label2id = {v:k for k,v in id2label.items()}
general_tags = df[df["category"] == 0].index
character_tags = df[df["category"] == 4].index
all_tags = df.index

tag_pair_df = pd.read_parquet("hf://datasets/p1atdev/danbooru-ja-tag-pair-20241015/data/train-00000-of-00001.parquet")
tag_pair = {title:other_names[0] for title, other_names in zip(tag_pair_df["title"], tag_pair_df["other_names"])}
for tag in df["name"]:
    if tag not in tag_pair:
        tag_pair[tag] = ""

def predict(target_tags, search_in):
    target_tags = [tag.strip().replace(" ", "_") for tag in target_tags.split(",")]
    target_ids = [label2id[tag] for tag in target_tags]
    query = head[target_ids].unsqueeze(1)

    sim = torch.cosine_similarity(query, head.unsqueeze(0), dim=2).mean(dim=0)
    tags = general_tags if search_in == "general" else character_tags if search_in == "character" else all_tags
    return {f"{id2label[i]}({tag_pair[id2label[i]]})": sim[i].item() for i in tags}

demo = gr.Interface(
        fn=predict, 
        inputs=[
            gr.Text(value="pink hair, braid", label="Target tags"),
            gr.Dropdown(["all", "general", "character"], label="Search in", value="all")
        ],
        outputs=gr.Label(num_top_classes=50),
        title=TITLE,
        description=DESCRIPTION
    )

demo.launch()