Add files using upload-large-folder tool

annotator/annotate.py

@@ -0,0 +1,237 @@
+import os
+import json
+from datetime import datetime
+from utils import divide_prompt
+from flask import Flask, render_template, request, send_from_directory, jsonify
+
+app = Flask(__name__)
+
+# 设置图像目录路径
+IMAGE_DIR = "../images_divided"
+ITEMS_PER_PAGE = 1  # 每页显示的 data 项目数
+DATA_FILE = "data/annotation.jsonl"  # 保存排序结果的文件
+PROMPT_DATA_FILE = 'data/extended_data_summary_v2.jsonl'
+
+with open(PROMPT_DATA_FILE, 'r', encoding='utf-8') as f:
+    PROMPT_DATA = [json.loads(line) for line in f if line.strip()]
+    PROMPT_DATA = {item['idx']: item for item in PROMPT_DATA}
+
+@app.route('/')
+def index():
+    # 读取所有文件夹
+    folders = sorted([
+        f for f in os.listdir(IMAGE_DIR)
+        if os.path.isdir(os.path.join(IMAGE_DIR, f)) and not f.startswith('.')
+    ])
+    data = []
+    
+    # 加载已有的标注记录
+    existing_records = load_existing_records()
+    existing_dict = {}
+    for record in existing_records:
+        folder = record.get('folder')
+        ref_image = record.get('ref_image')
+        compare_images = tuple(sorted(record.get('compare_images', [])))  # 使用排序后的元组作为键的一部分
+        key = (folder, ref_image, compare_images)
+        existing_dict[key] = {
+            'compare_images': record.get('compare_images', []),
+            'rank_images': record.get('rank_images', [])
+        }
+    
+    for cnt, idx in enumerate(folders):
+        folder_path = os.path.join(IMAGE_DIR, idx)
+        images = sorted([img for img in os.listdir(folder_path) if img.endswith('.png')])
+        first_indices, second_indices = zip(*[
+            list(map(int, img.split('.')[0].split('_')))
+            for img in images
+        ])
+        first_indices = sorted(set(first_indices))
+        second_indices = sorted(set(second_indices))
+        
+        for i in first_indices:
+            for j in second_indices:
+                ref_img = f"{i}_{j}.png"
+                candidates = [
+                    [
+                        f"{x}_{y}.png"
+                        for x in first_indices
+                    ]
+                    for y in second_indices if y != j
+                ]
+                
+                items = [
+                    {
+                        'ref_image': f"/images/{idx}/{ref_img}",
+                        'compare_images': [f"/images/{idx}/{cand_img}" for cand_img in cand],  # 待比较的图像列表
+                        'folder': idx,
+                        'ref_index': (i, j),
+                        'candidate_column': y,
+                        'prompt': divide_prompt(PROMPT_DATA[idx]['prompt'])[0] if idx in PROMPT_DATA and 'prompt' in PROMPT_DATA[idx] else '',
+                        'existing_compare_images': [],  # 临时占位，后面会更新
+                        'existing_rank_images': [],  # 临时占位，后面会更新
+                        'is_annotated': False,  # 临时占位，后面会更新
+                    } for y, cand in zip([y for y in second_indices if y != j], candidates)
+                ]
+                
+                # 对每个 item 检查是否已有标注
+                for item in items:
+                    # 从完整路径中提取文件名
+                    current_compare_images = [img.split('/')[-1] for img in item['compare_images']]
+                    
+                    # 检查是否已有标注（使用 ref_image 和 compare_images 一起判定）
+                    ref_filename = item['ref_image'].split('/')[-1]
+                    lookup_key = (idx, ref_filename, tuple(sorted(current_compare_images)))
+                    existing_data = existing_dict.get(lookup_key, {})
+                    
+                    # 更新 item 的标注信息
+                    item['existing_compare_images'] = existing_data.get('compare_images', [])
+                    item['existing_rank_images'] = existing_data.get('rank_images', [])
+                    item['is_annotated'] = len(existing_data.get('rank_images', [])) > 0
+                
+                data.extend(items)
+
+    # 分页逻辑
+    page = int(request.args.get('page', 1))
+    start = (page - 1) * ITEMS_PER_PAGE
+    end = start + ITEMS_PER_PAGE
+    paginated_data = data[start:end]
+
+    return render_template(
+        'index.html', 
+        data=paginated_data, 
+        page=page, 
+        total_pages=(len(data) + ITEMS_PER_PAGE - 1) // ITEMS_PER_PAGE
+    )
+
+@app.route('/sort', methods=['POST'])
+def sort_images():
+    try:
+        # 获取排序后的数据
+        sorted_images = request.form.getlist('sorted_images')
+        
+        if not sorted_images:
+            return jsonify({'status': 'error', 'message': '没有排序数据'}), 400
+        
+        # 解析排序数据
+        ranking = []
+        for item in sorted_images:
+            rank, image_path = item.split(':', 1)
+            ranking.append({
+                'rank': int(rank),
+                'image_path': image_path
+            })
+        
+        # 排序以确保顺序正确
+        ranking.sort(key=lambda x: x['rank'])
+        
+        # 提取有用信息
+        if ranking:
+            first_image = ranking[0]['image_path']
+            parts = first_image.split('/')
+            folder = parts[2]  # folder_name
+            
+            # 获取参考图像信息（从表单中）
+            ref_image = request.form.get('ref_image', '')
+            
+            # 解析参考图像文件名
+            ref_filename = ref_image.split('/')[-1]  # x_y.png
+            
+            # 获取原始的 compare_images（所有候选图像的原始顺序）
+            all_compare_images = request.form.get('all_compare_images', '')
+            if all_compare_images:
+                all_compare_images = json.loads(all_compare_images)
+            else:
+                all_compare_images = []
+            
+            # 解析排序图像的文件名 - 这是用户的排序结果
+            ranked_filenames = []
+            for item in ranking:
+                filename = item['image_path'].split('/')[-1]
+                ranked_filenames.append(filename)
+            
+            # 创建唯一的数据记录
+            record = {
+                'folder': folder,
+                'ref_image': ref_filename,
+                'compare_images': all_compare_images,  # 原始顺序的所有候选图像
+                'rank_images': ranked_filenames,  # 用户的排序结果
+                'timestamp': datetime.now().isoformat()
+            }
+            
+            # 读取现有数据以检查唯一性
+            existing_records = load_existing_records()
+            
+            # 检查是否已存在相同的记录（基于 folder, ref_image 和 compare_images）
+            updated = False
+            for i, existing in enumerate(existing_records):
+                existing_compare_sorted = tuple(sorted(existing.get('compare_images', [])))
+                current_compare_sorted = tuple(sorted(all_compare_images))
+                
+                if (existing.get('folder') == folder and 
+                    existing.get('ref_image') == ref_filename and
+                    existing_compare_sorted == current_compare_sorted):
+                    # 更新现有记录
+                    existing_records[i] = record
+                    updated = True
+                    break
+            
+            if not updated:
+                # 添加新记录
+                existing_records.append(record)
+            
+            # 保存到文件
+            save_records(existing_records)
+            
+            print(f"{'更新' if updated else '添加'}排序记录:", record)
+            
+            return jsonify({
+                'status': 'success', 
+                'message': f"排序已{'更新' if updated else '保存'}！",
+                'record': record
+            })
+    
+    except Exception as e:
+        print(f"错误: {str(e)}")
+        import traceback
+        traceback.print_exc()
+        return jsonify({'status': 'error', 'message': str(e)}), 500
+
+def load_existing_records():
+    """加载现有的记录"""
+    if not os.path.exists(DATA_FILE):
+        return []
+    
+    records = []
+    with open(DATA_FILE, 'r', encoding='utf-8') as f:
+        for line in f:
+            line = line.strip()
+            if line:
+                records.append(json.loads(line))
+    return records
+
+def save_records(records):
+    """保存记录到JSONL文件"""
+    # 确保目录存在
+    os.makedirs(os.path.dirname(DATA_FILE), exist_ok=True)
+    
+    with open(DATA_FILE, 'w', encoding='utf-8') as f:
+        for record in records:
+            f.write(json.dumps(record, ensure_ascii=False) + '\n')
+
+# 添加一个路由来提供图片文件
+@app.route('/images/<path:filename>')
+def serve_images(filename):
+    return send_from_directory(IMAGE_DIR, filename)
+
+# 添加一个查看已保存数据的路由
+@app.route('/view_data')
+def view_data():
+    """查看已保存的排序数据"""
+    records = load_existing_records()
+    return jsonify({
+        'total': len(records),
+        'records': records
+    })
+
+if __name__ == '__main__':
+    app.run(debug=True)

annotator/templates/index.html

@@ -0,0 +1,255 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>Image Ranking</title>
+    <style>
+        body {
+            font-family: Arial, sans-serif;
+            padding: 20px;
+        }
+
+        .rank-btn {
+            padding: 5px 15px;
+            cursor: pointer;
+            border: 2px solid #4CAF50;
+            background-color: white;
+            color: #4CAF50;
+            border-radius: 4px;
+            font-weight: bold;
+            min-width: 60px;
+        }
+        .rank-btn.ranked {
+            background-color: #4CAF50;
+            color: white;
+            border-color: #4CAF50;
+        }
+        .rank-btn:hover {
+            opacity: 0.8;
+        }
+        .submit-btn {
+            margin-top: 20px;
+            padding: 10px 30px;
+            background-color: #2196F3;
+            color: white;
+            border: none;
+            border-radius: 4px;
+            cursor: pointer;
+            font-size: 16px;
+        }
+        .submit-btn:hover {
+            background-color: #0b7dda;
+        }
+        .submit-btn:disabled {
+            background-color: #cccccc;
+            cursor: not-allowed;
+        }
+        .show-annotation-btn {
+            margin-top: 10px;
+            padding: 8px 20px;
+            background-color: #FF9800;
+            color: white;
+            border: none;
+            border-radius: 4px;
+            cursor: pointer;
+            font-size: 14px;
+        }
+        .show-annotation-btn:hover {
+            background-color: #F57C00;
+        }
+        .annotation-status {
+            display: inline-block;
+            margin-left: 15px;
+            padding: 5px 12px;
+            background-color: #4CAF50;
+            color: white;
+            border-radius: 4px;
+            font-size: 14px;
+            font-weight: bold;
+        }
+        /* 消息提示样式 */
+        .message-box {
+            position: fixed;
+            top: 20px;
+            right: 20px;
+            padding: 15px 25px;
+            border-radius: 4px;
+            font-size: 16px;
+            font-weight: bold;
+            box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1);
+            z-index: 1000;
+            display: none;
+            animation: slideIn 0.3s ease-out;
+        }
+        .message-box.success {
+            background-color: #4CAF50;
+            color: white;
+        }
+        .message-box.error {
+            background-color: #f44336;
+            color: white;
+        }
+        .message-box.show {
+            display: block;
+        }
+        @keyframes slideIn {
+            from {
+                transform: translateX(100%);
+                opacity: 0;
+            }
+            to {
+                transform: translateX(0);
+                opacity: 1;
+            }
+        }
+        @keyframes slideOut {
+            from {
+                transform: translateX(0);
+                opacity: 1;
+            }
+            to {
+                transform: translateX(100%);
+                opacity: 0;
+            }
+        }
+        .message-box.hiding {
+            animation: slideOut 0.3s ease-out;
+        }
+
+        /* 固定在右上角的分页条 */
+        .page-control {
+            position: fixed;
+            top: 20px;
+            right: 20px;
+            background: rgba(255,255,255,0.9);
+            border: 1px solid #ddd;
+            padding: 8px 12px;
+            border-radius: 6px;
+            box-shadow: 0 2px 6px rgba(0,0,0,0.1);
+            z-index: 999;
+            display: flex;
+            align-items: center;
+            gap: 8px;
+            font-size: 14px;
+        }
+        .page-control input {
+            width: 60px;
+            padding: 3px 5px;
+            border: 1px solid #ccc;
+            border-radius: 4px;
+        }
+        .page-control button {
+            padding: 4px 8px;
+            background-color: #2196F3;
+            color: white;
+            border: none;
+            border-radius: 4px;
+            cursor: pointer;
+        }
+        .page-control button:hover {
+            background-color: #0b7dda;
+        }
+    </style>
+    <script>
+        let rankCounter = 1;
+        const rankedImages = new Map(); 
+        let allCompareImages = []; 
+
+        function initializeCompareImages() {
+            allCompareImages = [];
+            const rankButtons = document.querySelectorAll('.rank-btn');
+            rankButtons.forEach(btn => {
+                const imagePath = btn.getAttribute('data-image');
+                const filename = imagePath.split('/').pop();
+                if (!allCompareImages.includes(filename)) {
+                    allCompareImages.push(filename);
+                }
+            });
+        }
+        window.addEventListener('DOMContentLoaded', function() {
+            initializeCompareImages();
+        });
+
+        /* 这里省略 rankImage、loadExistingAnnotation、showMessage、submitRanking 的原有实现 */
+        /* ... 你的原始脚本保持不变 ... */
+
+        // 新增跳转函数
+        function jumpToPage(event) {
+            event.preventDefault();
+            const input = document.getElementById('pageInput');
+            let page = parseInt(input.value);
+            if (isNaN(page) || page < 1) page = 1;
+            if (page > {{ total_pages }}) page = {{ total_pages }};
+            window.location.href = "/?page=" + page;
+            return false;
+        }
+    </script>
+</head>
+<body>
+    <!-- 消息提示框 -->
+    <div id="messageBox" class="message-box"></div>
+
+    <!-- 固定在右上角的分页控制条 -->
+    <div class="page-control">
+        <span>{{ page }}/{{ total_pages }}</span>
+        <form id="pageJumpForm" onsubmit="return jumpToPage(event)" style="display:inline;">
+            <input type="number" id="pageInput" min="1" max="{{ total_pages }}" value="{{ page }}">
+            <button type="submit">跳转</button>
+        </form>
+    </div>
+
+    <h1>Image Ranking</h1>
+    <form id="rankingForm" action="/sort" method="POST" onsubmit="return submitRanking(event)">
+        {% for item in data %}
+            <div>
+                {% if item.prompt %}
+                <div style="background-color: #f0f0f0; padding: 15px; margin-bottom: 20px; border-radius: 4px; border-left: 4px solid #2196F3;">
+                    <h2 style="margin-top: 0; color: #333;">Prompt:</h2>
+                    <p style="font-size: 16px; line-height: 1.6; margin-bottom: 0; color: #555;">{{ item.prompt }}</p>
+                </div>
+                {% endif %}
+                <h3>Reference Image: {{ item.ref_image }}
+                    {% if item.is_annotated %}
+                    <span class="annotation-status">已标注</span>
+                    {% endif %}
+                </h3>
+                <p>Folder: {{ item.folder }}</p>
+                <img src="{{ item.ref_image }}" alt="ref_image" width="200">
+                <input type="hidden" name="ref_image" value="{{ item.ref_image }}">
+                {% if item.is_annotated %}
+                <button type="button" class="show-annotation-btn" 
+                        data-loaded="false"
+                        onclick='loadExistingAnnotation(this, "{{ item.folder }}", "{{ item.ref_image.split("/")[-1] }}", {{ item.existing_rank_images|tojson|safe }})'>
+                    显示已有标注
+                </button>
+                {% endif %}
+                <h4>Compare Images:</h4>
+                <div style="display: flex; gap: 15px; flex-wrap: wrap; align-items: flex-start;">
+                    {% for compare_image in item.compare_images %}
+                        <div class="image-item" style="display: flex; flex-direction: column; align-items: center; gap: 5px;">
+                            <img src="{{ compare_image }}" alt="compare_image" width="100">
+                            <button type="button" 
+                                    class="rank-btn" 
+                                    data-image="{{ compare_image }}"
+                                    onclick="rankImage(this, '{{ compare_image }}')">
+                                排序
+                            </button>
+                        </div>
+                    {% endfor %}
+                </div>
+            </div>
+        {% endfor %}
+        <button type="submit" class="submit-btn">提交排序</button>
+    </form>
+
+    <div style="margin-top: 20px;">
+        {% if page > 1 %}
+            <a href="/?page={{ page - 1 }}" style="margin-right: 10px;">上一页</a>
+        {% endif %}
+        {% if page < total_pages %}
+            <a href="/?page={{ page + 1 }}">下一页</a>
+        {% endif %}
+    </div>
+</body>
+</html>

annotator/test.ipynb

@@ -0,0 +1,130 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 16,
+   "id": "5bdfe87c",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "import json\n",
+    "from itertools import combinations\n",
+    "IMAGE_DIR = \"/root/siton-tmp/images_divided\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "fb8c9036",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "folders = sorted([\n",
+    "    f for f in os.listdir(IMAGE_DIR)\n",
+    "    if os.path.isdir(os.path.join(IMAGE_DIR, f)) and not f.startswith('.')\n",
+    "])\n",
+    "data = []\n",
+    "for cnt, idx in enumerate(folders):\n",
+    "    folder_path = os.path.join(IMAGE_DIR, idx)\n",
+    "    images = sorted([img for img in os.listdir(folder_path) if img.endswith('.png')])\n",
+    "    first_indices, second_indices = zip(*[\n",
+    "        list(map(int, img.split('.')[0].split('_')))\n",
+    "        for img in images\n",
+    "    ])\n",
+    "    first_indices = sorted(set(first_indices))\n",
+    "    second_indices = sorted(set(second_indices))\n",
+    "    \n",
+    "    for i in first_indices:\n",
+    "        for j in second_indices:\n",
+    "            ref_img = f\"{i}_{j}.png\"\n",
+    "            candidates = [\n",
+    "                [\n",
+    "                    f\"{x}_{y}.png\"\n",
+    "                    for x in first_indices\n",
+    "                ]\n",
+    "                for y in second_indices if y != j\n",
+    "            ]\n",
+    "            items = [\n",
+    "                {\n",
+    "                'ref_image': ref_img,\n",
+    "                'rank_images': cand,\n",
+    "                'idx': idx,\n",
+    "                } for cand in candidates\n",
+    "            ]\n",
+    "            data.extend(items)\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 19,
+   "id": "36e0603b",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "[{'ref_image': '0_0.png',\n",
+       "  'rank_images': ['0_1.png', '1_1.png', '2_1.png', '3_1.png'],\n",
+       "  'idx': '0000'},\n",
+       " {'ref_image': '0_0.png',\n",
+       "  'rank_images': ['0_2.png', '1_2.png', '2_2.png', '3_2.png'],\n",
+       "  'idx': '0000'},\n",
+       " {'ref_image': '0_1.png',\n",
+       "  'rank_images': ['0_0.png', '1_0.png', '2_0.png', '3_0.png'],\n",
+       "  'idx': '0000'},\n",
+       " {'ref_image': '0_1.png',\n",
+       "  'rank_images': ['0_2.png', '1_2.png', '2_2.png', '3_2.png'],\n",
+       "  'idx': '0000'},\n",
+       " {'ref_image': '0_2.png',\n",
+       "  'rank_images': ['0_0.png', '1_0.png', '2_0.png', '3_0.png'],\n",
+       "  'idx': '0000'},\n",
+       " {'ref_image': '0_2.png',\n",
+       "  'rank_images': ['0_1.png', '1_1.png', '2_1.png', '3_1.png'],\n",
+       "  'idx': '0000'},\n",
+       " {'ref_image': '1_0.png',\n",
+       "  'rank_images': ['0_1.png', '1_1.png', '2_1.png', '3_1.png'],\n",
+       "  'idx': '0000'},\n",
+       " {'ref_image': '1_0.png',\n",
+       "  'rank_images': ['0_2.png', '1_2.png', '2_2.png', '3_2.png'],\n",
+       "  'idx': '0000'},\n",
+       " {'ref_image': '1_1.png',\n",
+       "  'rank_images': ['0_0.png', '1_0.png', '2_0.png', '3_0.png'],\n",
+       "  'idx': '0000'},\n",
+       " {'ref_image': '1_1.png',\n",
+       "  'rank_images': ['0_2.png', '1_2.png', '2_2.png', '3_2.png'],\n",
+       "  'idx': '0000'}]"
+      ]
+     },
+     "execution_count": 19,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "data[:10]"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "pytorch-env",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.11"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

annotator/utils.py

@@ -0,0 +1,13 @@
+import re
+
+
+def divide_prompt(prompt):
+    # seqis like ". [TOP-LEFT]:"
+    match_sep = re.compile(r"\.\s+[A-Z0-9-\[\]]+:")
+    seps = match_sep.findall(prompt)
+    # Add '.' for each sentence
+    sub_prompts = [
+        p + '.' if p.strip()[-1] != '.' else p
+        for p in re.split('|'.join(map(re.escape, seps)), prompt)
+    ]
+    return sub_prompts