text2image.py

from functools import partial
import json
from multiprocessing.pool import ThreadPool as Pool
import gradio as gr
from utils import *

from clip_retrieval.clip_client import ClipClient


def text2image_gr():
    def clip_api(query_text='', return_n=8, model_name=clip_base, thumbnail="是"):
        #client = ClipClient(url="http://9.135.121.52:1234//knn-service",
        client = ClipClient(url="http://127.0.0.1:1234//knn-service",
                            indice_name="ltr_cover_index",
                            aesthetic_weight=0,
                            num_images=int(return_n))
        #result = client.query(embedding_input=query_emb)
        result = client.query(text=query_text)
        
        if not result or len(result) == 0:
            print("no result found")
            return None
        
        print(f"get result sucessed, num: {len(result)}")
        
        cover_urls = [res['cover_url'] for res in result]
        cover_info = []
        for res in result:
            json_info = {"cover_url": res['cover_url'],
                         "similarity": round(res['similarity'], 6),
                         "docid": res['docids']}
            cover_info.append(str(json_info))
        pool = Pool()
        new_url2image = partial(url2img, thumbnail=thumbnail)
        ret_imgs = pool.map(new_url2image, cover_urls)
        pool.close()
        pool.join()

        new_ret = []
        for i in range(len(ret_imgs)):
            new_ret.append([ret_imgs[i], cover_info[i]])
        return new_ret

    examples = [
        ["cat", 12, clip_base, "是"],
        ["dog", 12, clip_base, "是"],
        ["bag", 12, clip_base, "是"],
        ["a cat is sit on the table", 12, clip_base, "是"]
    ]

    title = "<h1 align='center'>CLIP文到图搜索应用</h1>"

    with gr.Blocks() as demo:
        gr.Markdown(title)
        gr.Markdown(description)
        with gr.Row():
            with gr.Column(scale=1):
                with gr.Column(scale=2):
                    text = gr.Textbox(value="cat", label="请填写文本", elem_id=0, interactive=True)
                num = gr.components.Slider(minimum=0, maximum=50, step=1, value=8, label="返回图片数（可能被过滤部分）", elem_id=2)
                model = gr.components.Radio(label="模型选择", choices=[clip_base],
                                            value=clip_base, elem_id=3)
                thumbnail = gr.components.Radio(label="是否返回缩略图", choices=[yes, no],
                                                value=yes, elem_id=4)
                btn = gr.Button("搜索", )
            with gr.Column(scale=100):
                out = gr.Gallery(label="检索结果为：", columns=4, height="auto") #.style(grid=4, height=200)
        inputs = [text, num, model, thumbnail]
        btn.click(fn=clip_api, inputs=inputs, outputs=out)
        gr.Examples(examples, inputs=inputs)
    return demo

if __name__ == "__main__":
    gr.close_all()
    with gr.TabbedInterface(
            [text2image_gr()],
            ["文到图搜索"],
    ) as demo:
        demo.launch(server_name='127.0.0.1', share=False)