viewer.py

import base64
import io
from PIL import Image
import gradio as gr
import pandas as pd

def decode_image(base64_str):
    try:
        img_data = base64.b64decode(base64_str)
        img = Image.open(io.BytesIO(img_data))
        return img
    except:
        return None

def display_sample(df, idx):
    sample = df.iloc[idx]
    
    title = sample['title']
    description = sample['description']
    label = "广告" if sample['label'] == 1 else "非广告"
    date = sample['date']
    comments = sample['comments'] if 'comments' in sample else []
    
    images = []
    for img_b64 in sample['images']:
        img = decode_image(img_b64)
        if img:
            images.append(img)
    
    return title, description, label, date, comments, images

def create_demo(df):
    with gr.Blocks() as demo:
        gr.Markdown("# RedNote Covert Advertisement Detection Dataset Viewer")
        
        with gr.Row():
            with gr.Column(scale=1):
                idx_slider = gr.Slider(minimum=0, maximum=len(df)-1, step=1, value=0, label="Sample Index")
                label_filter = gr.Radio(["全部", "仅广告", "仅非广告"], value="全部", label="筛选")
                
                def update_slider(choice):
                    if choice == "仅广告":
                        ad_indices = df[df['label'] == 1].index.tolist()
                        return gr.Slider(minimum=0, maximum=len(ad_indices)-1, step=1, value=0)
                    elif choice == "仅非广告":
                        non_ad_indices = df[df['label'] == 0].index.tolist()
                        return gr.Slider(minimum=0, maximum=len(non_ad_indices)-1, step=1, value=0)
                    else:
                        return gr.Slider(minimum=0, maximum=len(df)-1, step=1, value=0)
                
                label_filter.change(update_slider, inputs=[label_filter], outputs=[idx_slider])
            
            with gr.Column(scale=3):
                title_text = gr.Textbox(label="标题")
                desc_text = gr.Textbox(label="描述", lines=5)
                label_text = gr.Textbox(label="标签")
                date_text = gr.Textbox(label="日期")
                comments_text = gr.Textbox(label="评论", lines=5)
                image_gallery = gr.Gallery(label="图片", columns=3, height=400)
        
        def get_filtered_index(idx, filter_choice):
            if filter_choice == "仅广告":
                ad_indices = df[df['label'] == 1].index.tolist()
                return ad_indices[idx]
            elif filter_choice == "仅非广告":
                non_ad_indices = df[df['label'] == 0].index.tolist()
                return non_ad_indices[idx]
            else:
                return idx
        
        def update_display(idx, filter_choice):
            real_idx = get_filtered_index(idx, filter_choice)
            return display_sample(df, real_idx)
        
        idx_slider.change(
            update_display, 
            inputs=[idx_slider, label_filter], 
            outputs=[title_text, desc_text, label_text, date_text, comments_text, image_gallery]
        )
        
        # 初始显示第一个样本
        title, desc, label, date, comments, images = display_sample(df, 0)
        title_text.value = title
        desc_text.value = desc
        label_text.value = label
        date_text.value = date
        comments_text.value = "\n".join(comments) if comments else ""
        image_gallery.value = images
    
    return demo

# 加载数据集
def load_dataset():
    try:
        train_df = pd.read_parquet("train.parquet")
        val_df = pd.read_parquet("validation.parquet")
        test_df = pd.read_parquet("test.parquet")
        return pd.concat([train_df, val_df, test_df])
    except:
        # 如果加载失败，返回一个示例数据框
        return pd.DataFrame({
            'title': ['示例标题'],
            'description': ['这是一个示例描述'],
            'label': [0],
            'date': ['01-01'],
            'comments': [['评论1', '评论2']],
            'images': [['']]  # 空图片
        })

# 创建演示
df = load_dataset()
demo = create_demo(df)