app.py

from turtle import title
import gradio as gr
from transformers import pipeline
import numpy as np
from PIL import Image


pipes = {
    "openAI-ViT/B-16": pipeline("zero-shot-image-classification", model="openai/clip-vit-base-patch16"),
    "ViT/B-16": pipeline("zero-shot-image-classification", model="OFA-Sys/chinese-clip-vit-base-patch16"),
    "ViT/L-14": pipeline("zero-shot-image-classification", model="OFA-Sys/chinese-clip-vit-large-patch14"),
    "ViT/L-14@336px": pipeline("zero-shot-image-classification", model="OFA-Sys/chinese-clip-vit-large-patch14-336px"),
    "ViT/H-14": pipeline("zero-shot-image-classification", model="OFA-Sys/chinese-clip-vit-huge-patch14"),
}
inputs = [
    gr.inputs.Image(type='pil', 
                    label="Image 输入图片"),
    gr.inputs.Textbox(lines=1, 
                      label="Candidate Labels 候选分类标签"),
    gr.inputs.Radio(choices=[   
                                "openAI-ViT/B-16"
                                "ViT/B-16",
                                "ViT/L-14", 
                                "ViT/L-14@336px", 
                                "ViT/H-14",
                            ], type="value", default="ViT/B-16", label="Model 模型规模"), 
    gr.inputs.Textbox(lines=1, 
                      label="Prompt Template Prompt模板 ({}指代候选标签)", 
                      default="一张{}的图片。"),
]
images="festival.jpg"

def shot(image, labels_text, model_name, hypothesis_template):
    labels = [label.strip(" ") for label in labels_text.strip(" ").split(",")]
    res = pipes[model_name](images=image, 
           candidate_labels=labels,
           hypothesis_template=hypothesis_template)
    return {dic["label"]: dic["score"] for dic in res}

lei = "机动车道,非机动车道,人车混行道路,斑马线人行道,主干道路,乡间道路,内部小巷,人行横道,十字路口,丁字路口,岔路口,铁路沿线,铁路路口,高架桥,立交桥,过街天桥,桥梁,天桥上下口,地下隧道,地下人行通道,隧道通行区域,穿山隧道,隧道出入口,水池,河流,湖面,室外停车场,路面划线停车位,城市广场,裸露农田,林区,草坪,树木,公交站台,收费站,检查站,加油站,岗亭,车行道闸,人行闸机,安检机器,铁门,保安亭,门或电动门,人员出入口,车辆出入口,广告牌,横幅,沿街商铺,露天烧烤摊,超市,建筑施工,道路施工,人员卡口,车辆卡口,人行闸机,场所主出入口,安检门,X光安检机,电梯内部,扶梯,楼梯,台阶,室内通道,走廊,前台区域,公共大厅,室内停车场"

iface = gr.Interface(shot, 
            inputs, 
            "label", 
            examples=[["street.jpg", lei, "ViT/B-16", "一张{}的图片。"]],
            description="""<p>Chinese CLIP is a contrastive-learning-based vision-language foundation model pretrained on large-scale Chinese data. For more information, please refer to the paper and official github. Also, Chinese CLIP has already been merged into Huggingface Transformers! <br><br>
            Paper: <a href='https://arxiv.org/abs/2211.01335'>https://arxiv.org/abs/2211.01335</a> <br>
            Github: <a href='https://github.com/OFA-Sys/Chinese-CLIP'>https://github.com/OFA-Sys/Chinese-CLIP</a> (Welcome to star! 🔥🔥) <br><br>
            To play with this demo, add a picture and a list of labels in Chinese separated by commas. 上传图片，并输入多个分类标签，用英文逗号分隔。可点击页面最下方示例参考。<br>
            You can duplicate this space and run it privately: <a href='https://huggingface.co/spaces/OFA-Sys/chinese-clip-zero-shot-image-classification?duplicate=true'><img src='https://img.shields.io/badge/-Duplicate%20Space-blue?labelColor=white&style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAAAXNSR0IArs4c6QAAAP5JREFUOE+lk7FqAkEURY+ltunEgFXS2sZGIbXfEPdLlnxJyDdYB62sbbUKpLbVNhyYFzbrrA74YJlh9r079973psed0cvUD4A+4HoCjsA85X0Dfn/RBLBgBDxnQPfAEJgBY+A9gALA4tcbamSzS4xq4FOQAJgCDwV2CPKV8tZAJcAjMMkUe1vX+U+SMhfAJEHasQIWmXNN3abzDwHUrgcRGmYcgKe0bxrblHEB4E/pndMazNpSZGcsZdBlYJcEL9Afo75molJyM2FxmPgmgPqlWNLGfwZGG6UiyEvLzHYDmoPkDDiNm9JR9uboiONcBXrpY1qmgs21x1QwyZcpvxt9NS09PlsPAAAAAElFTkSuQmCC&logoWidth=14' alt='Duplicate Space'></a></p>""",
            title="Zero-shot Image Classification (中文零样本图像分类)")

iface.launch()