vis_conversation.py

import json
import random
from pathlib import Path
import matplotlib.pyplot as plt
import matplotlib.patches as patches
from PIL import Image
import numpy as np

def scale_bbox_to_1023(bbox, img_width, img_height):
    """将 bbox 坐标缩放到 0-1023 范围"""
    x1, y1, x2, y2 = bbox
    x1_scaled = int(x1 / img_width * 1023)
    y1_scaled = int(y1 / img_height * 1023)
    x2_scaled = int(x2 / img_width * 1023)
    y2_scaled = int(y2 / img_height * 1023)
    return [x1_scaled, y1_scaled, x2_scaled, y2_scaled]

def visualize_fewshot_sample(image_path, prompt_boxes, all_boxes, category, output_path, img_width=None, img_height=None):
    """
    可视化 few-shot 样本
    - prompt_boxes: 蓝色 (prompt 框)
    - all_boxes: 红色 (所有框)
    """
    # 读取图片
    img = Image.open(image_path)
    if img_width and img_height:
        # 如果需要缩放显示，可以调整
        img = img.resize((img_width, img_height))
    
    fig, ax = plt.subplots(1, figsize=(12, 9))
    ax.imshow(img)
    
    # 记录已画的框，用于处理重叠
    drawn_boxes = []
    
    # 先画所有框（红色，较粗，带透明度）
    for box_info in all_boxes:
        bbox = box_info['bbox']
        class_id = box_info.get('class_id', 0)
        
        # 创建矩形
        rect = patches.Rectangle(
            (bbox[0], bbox[1]), 
            bbox[2] - bbox[0], 
            bbox[3] - bbox[1],
            linewidth=2, 
            edgecolor='red', 
            facecolor='none',
            alpha=0.8
        )
        ax.add_patch(rect)
        
        # 添加标签（在框的左上角）
        label = f"{category}_{class_id}"
        ax.text(
            bbox[0], bbox[1] - 5, 
            label, 
            fontsize=8, 
            color='red', 
            weight='bold',
            bbox=dict(boxstyle='round,pad=0.3', facecolor='white', alpha=0.7, edgecolor='red')
        )
        drawn_boxes.append(('red', bbox))
    
    # 再画 prompt 框（蓝色，虚线，更突出）
    for box_info in prompt_boxes:
        bbox = box_info['bbox']
        class_id = box_info.get('class_id', 0)
        
        # 创建矩形（蓝色虚线）
        rect = patches.Rectangle(
            (bbox[0], bbox[1]), 
            bbox[2] - bbox[0], 
            bbox[3] - bbox[1],
            linewidth=3, 
            edgecolor='blue', 
            facecolor='none',
            linestyle='--',
            alpha=0.9
        )
        ax.add_patch(rect)
        
        # 添加标签（在框的右上角）
        label = f"PROMPT_{category}_{class_id}"
        ax.text(
            bbox[2], bbox[1] - 5, 
            label, 
            fontsize=8, 
            color='blue', 
            weight='bold',
            bbox=dict(boxstyle='round,pad=0.3', facecolor='white', alpha=0.7, edgecolor='blue'),
            horizontalalignment='right'
        )
    
    # 添加图例
    from matplotlib.patches import Patch
    legend_elements = [
        Patch(facecolor='none', edgecolor='blue', linewidth=3, linestyle='--', label='Prompt Boxes (Few-shot)'),
        Patch(facecolor='none', edgecolor='red', linewidth=2, label='All GT Boxes')
    ]
    ax.legend(handles=legend_elements, loc='upper right', fontsize=10)
    
    # 添加标题
    ax.set_title(f"Few-shot Detection (k={len(prompt_boxes)} prompts)", fontsize=14, weight='bold')
    ax.axis('off')
    
    plt.tight_layout()
    plt.savefig(output_path, dpi=150, bbox_inches='tight')
    plt.close()
    print(f"  保存可视化: {output_path}")

def visualize_multiple_samples(data_file, output_dir, num_samples=10):
    """
    可视化多个样本
    """
    output_dir = Path(output_dir)
    output_dir.mkdir(parents=True, exist_ok=True)
    
    # 读取所有数据
    samples = []
    with open(data_file, 'r', encoding='utf-8') as f:
        for line in f:
            samples.append(json.loads(line.strip()))
    
    # 随机选择 num_samples 个样本
    if len(samples) > num_samples:
        selected_samples = random.sample(samples, num_samples)
    else:
        selected_samples = samples
    
    print(f"从 {len(samples)} 个样本中选择了 {len(selected_samples)} 个进行可视化")
    
    for idx, data in enumerate(selected_samples):
        image_path = data['image']
        prompt_boxes = data['prompt_boxes']
        all_boxes = data['all_boxes']
        category = all_boxes[0]['category'] if all_boxes else 'unknown'
        k = data['k']
        
        # 输出文件路径
        output_path = output_dir / f"sample_{idx+1}_k{k}.png"
        
        print(f"\n可视化样本 {idx+1}:")
        print(f"  图片: {Path(image_path).name}")
        print(f"  K={k}, Prompt框数={len(prompt_boxes)}, 总框数={len(all_boxes)}")
        print(f"  类别: {category}")
        
        try:
            visualize_fewshot_sample(
                image_path, 
                prompt_boxes, 
                all_boxes, 
                category, 
                output_path
            )
        except Exception as e:
            print(f"  错误: {e}")
    
    print(f"\n可视化完成！保存到: {output_dir}")

def main():
    # 设置随机种子
    random.seed(42)
    
    # 输入文件（k=2 的数据）
    input_file = Path("/home/disk2/hjl/ICL_QWEN/ICL_benchmark/fewshot_data/nested/fewshot_k2_nested.jsonl")
    
    # 输出目录
    output_dir = Path("/home/disk2/hjl/ICL_QWEN/ICL_benchmark/visualizations_k2")
    
    print("="*60)
    print("Few-shot 可视化工具 (k=2)")
    print("="*60)
    print(f"输入文件: {input_file}")
    print(f"输出目录: {output_dir}")
    print()
    
    # 可视化 10 个样本
    visualize_multiple_samples(input_file, output_dir, num_samples=10)

if __name__ == "__main__":
    main()