ZoneMaestro_code / eval /respace /debug_sample_issue.py
kkkkiiii's picture
Add files using upload-large-folder tool
ebc85b3 verified
Raw
History Blame Contribute Delete
15.2 kB
#!/usr/bin/env python3
"""
调试 sample.py 中的 'NoneType' object has no attribute 'get' 错误
专门针对 sample_all_assets 方法中的问题
"""
import json
import sys
import os
import traceback
import copy
import uuid
from pathlib import Path
# 添加 src 路径
sys.path.insert(0, '/home/v-meiszhang/amlt-project/respace')
def load_test_scene():
"""加载测试场景数据"""
json_path = "/home/v-meiszhang/amlt-project/group-layout/infer_results/individual_samples/full_20250831_112415/checkpoint-270/ae8d145b-5c7c-4970-a4b6-7855e64ea4eb-1f704044-378d-41ca-8978-43c0881b103c.json"
print(f"📖 加载测试场景: {json_path}")
with open(json_path, 'r', encoding='utf-8') as f:
raw_data = json.load(f)
# 提取和处理场景数据(简化版本)
if "predict" in raw_data:
scene_data = raw_data["predict"]
# 简单的 JSON 解析
if isinstance(scene_data, str):
import re
pattern = r'<answer>(.*?)</answer>'
match = re.search(pattern, scene_data, re.DOTALL)
if match:
scene_data = json.loads(match.group(1).strip())
else:
scene_data = json.loads(scene_data)
else:
scene_data = raw_data
# 处理房间边界
if "room_envelope" in scene_data:
envelope_data = scene_data["room_envelope"]
if isinstance(envelope_data, dict):
scene_data["bounds_top"] = envelope_data.get("bounds_top")
scene_data["bounds_bottom"] = envelope_data.get("bounds_bottom")
# 展平对象
if "groups" in scene_data:
scene_data["objects"] = []
for group in scene_data["groups"]:
if isinstance(group, dict) and "objects" in group:
scene_data["objects"].extend(group["objects"])
return scene_data
def debug_sample_all_assets():
"""调试 sample_all_assets 方法"""
print("🔬 开始调试 sample_all_assets 方法")
print("="*60)
# 加载环境变量
try:
from dotenv import load_dotenv
load_dotenv("/home/v-meiszhang/amlt-project/respace/.env")
print("✅ 环境变量加载成功")
except Exception as e:
print(f"⚠️ 环境变量加载失败: {e}")
# 加载测试场景
try:
scene = load_test_scene()
print(f"✅ 测试场景加载成功")
print(f" 对象数量: {len(scene.get('objects', []))}")
# 检查对象完整性
objects = scene.get('objects', [])
for i, obj in enumerate(objects):
if obj is None:
print(f" ❌ 对象 {i+1} 为 None")
elif not isinstance(obj, dict):
print(f" ❌ 对象 {i+1} 不是字典: {type(obj)}")
else:
desc = obj.get("desc", "")
size = obj.get("size", [])
print(f" ✅ 对象 {i+1}: desc='{desc[:30]}...', size={size}")
except Exception as e:
print(f"❌ 测试场景加载失败: {e}")
traceback.print_exc()
return
# 初始化采样引擎
try:
from src.sample import AssetRetrievalModule
print("\n🔧 初始化 AssetRetrievalModule...")
sampling_engine = AssetRetrievalModule(
lambd=0.5,
sigma=0.05,
temp=0.2,
top_p=0.95,
top_k=20,
asset_size_threshold=0.5,
rand_seed=1234,
dvc='cuda' if os.getenv('CUDA_VISIBLE_DEVICES') else 'cpu',
do_print=True # 启用详细输出
)
print("✅ AssetRetrievalModule 初始化成功")
# 检查元数据
print(f"\n📊 元数据统计:")
print(f" all_assets_metadata: {len(sampling_engine.all_assets_metadata)} 条目")
print(f" all_assets_metadata_scaled: {len(sampling_engine.all_assets_metadata_scaled)} 条目")
print(f" all_jids_catalog: {len(sampling_engine.all_jids_catalog)} 条目")
except Exception as e:
print(f"❌ AssetRetrievalModule 初始化失败: {e}")
traceback.print_exc()
return
# 修补 create_sampled_obj 方法以添加调试信息
original_create_sampled_obj = sampling_engine.create_sampled_obj
def debug_create_sampled_obj(obj, probs, is_greedy_sampling):
"""带调试信息的 create_sampled_obj"""
print(f"\n🎯 调试 create_sampled_obj:")
print(f" 输入对象类型: {type(obj)}")
if obj is None:
print(" ❌ 输入对象为 None!")
return None
print(f" 对象描述: {obj.get('desc', 'N/A')[:50]}...")
print(f" 对象大小: {obj.get('size', 'N/A')}")
try:
# 获取采样 jid
if obj.get("jid") == None:
import torch
if is_greedy_sampling:
_, idx_sampled = torch.max(probs, dim=0)
else:
idx_sampled = torch.multinomial(probs, num_samples=1)
jid_sampled_obj = sampling_engine.all_jids_catalog[idx_sampled]
else:
jid_sampled_obj = obj.get("jid")
print(f" 采样的 JID: {jid_sampled_obj}")
# 检查资产存在性
asset = sampling_engine.all_assets_metadata.get(jid_sampled_obj)
print(f" 在 all_assets_metadata 中: {asset is not None}")
if asset == None:
asset = sampling_engine.all_assets_metadata_scaled.get(jid_sampled_obj)
print(f" 在 all_assets_metadata_scaled 中: {asset is not None}")
if asset is None:
print(f" ❌ 无法找到 JID {jid_sampled_obj} 对应的资产!")
return None
# 检查缩放资产的完整性
size_sampled_obj = asset.get("size")
orig_jid = asset.get("jid")
print(f" 缩放资产大小: {size_sampled_obj}")
print(f" 原始 JID: {orig_jid}")
if orig_jid is None:
print(f" ❌ 缩放资产没有原始 JID!")
return None
orig_asset = sampling_engine.all_assets_metadata.get(orig_jid)
print(f" 原始资产存在: {orig_asset is not None}")
if orig_asset is None:
print(f" ❌ 无法找到原始资产 {orig_jid}!")
return None
desc_sampled_obj = orig_asset.get("summary")
print(f" 原始资产描述: {desc_sampled_obj[:30] if desc_sampled_obj else 'None'}...")
else:
desc_sampled_obj = asset.get("summary")
size_sampled_obj = asset.get("size")
print(f" 直接资产描述: {desc_sampled_obj[:30] if desc_sampled_obj else 'None'}...")
print(f" 直接资产大小: {size_sampled_obj}")
# 检查必要字段
if desc_sampled_obj is None:
print(f" ❌ 资产描述为 None!")
return None
if size_sampled_obj is None:
print(f" ❌ 资产大小为 None!")
return None
# 创建新对象
new_obj = copy.deepcopy(obj)
new_obj.update({
"sampled_asset_jid": jid_sampled_obj,
"sampled_asset_desc": desc_sampled_obj,
"sampled_asset_size": size_sampled_obj,
"uuid": str(uuid.uuid4())
})
print(f" ✅ 成功创建采样对象")
return new_obj
except Exception as e:
print(f" ❌ create_sampled_obj 内部错误: {e}")
traceback.print_exc()
return None # 临时替换方法
sampling_engine.create_sampled_obj = debug_create_sampled_obj
# 修补 sample_all_assets 方法以添加更多调试信息
print(f"\n🎯 开始调试 sample_all_assets 过程...")
try:
print(f"📊 场景信息:")
print(f" 对象总数: {len(scene.get('objects', []))}")
# 手动实现 sample_all_assets 的调试版本
batch_size = 4 # 小批次便于调试
sampled_scene = copy.deepcopy(scene)
sampled_scene["objects"] = []
desc_size_map = {}
objects = scene.get("objects", [])
descriptions = [obj.get("desc") for obj in objects]
sizes = [obj.get("size", []) for obj in objects]
print(f"📋 准备批处理:")
print(f" 描述数量: {len(descriptions)}")
print(f" 大小数量: {len(sizes)}")
print(f" 批大小: {batch_size}")
for batch_start in range(0, len(descriptions), batch_size):
batch_end = min(batch_start + batch_size, len(descriptions))
print(f"\n📦 处理批次 {batch_start}-{batch_end}")
batch_descriptions = descriptions[batch_start:batch_end]
batch_sizes = sizes[batch_start:batch_end]
print(f" 批次描述: {len(batch_descriptions)} 条")
print(f" 批次大小: {len(batch_sizes)} 条")
# 获取批次概率
try:
batch_probs = sampling_engine.forward_batch(batch_descriptions, batch_sizes)
print(f" ✅ 批次概率计算成功: {batch_probs.shape}")
except Exception as e:
print(f" ❌ 批次概率计算失败: {e}")
traceback.print_exc()
continue
# 处理批次中的每个对象
for i, obj in enumerate(objects[batch_start:batch_end]):
obj_idx = batch_start + i
print(f"\n🔍 处理对象 {obj_idx + 1}/{len(objects)}")
if obj is None:
print(f" ❌ 对象为 None,跳过")
continue
desc = obj.get("desc")
size = obj.get("size", [])
print(f" 描述: {desc[:30] if desc else 'None'}...")
print(f" 大小: {size}")
# 检查是否已有相同描述的对象
if desc in desc_size_map:
print(f" 🔍 在缓存中查找相似对象...")
matching_obj = None
for j, sampled_obj in enumerate(desc_size_map[desc]):
print(f" 检查缓存对象 {j+1}: {sampled_obj is not None}")
if sampled_obj is None:
print(f" ❌ 缓存对象 {j+1} 为 None!")
continue
if not isinstance(sampled_obj, dict):
print(f" ❌ 缓存对象 {j+1} 不是字典: {type(sampled_obj)}")
continue
if "size" not in sampled_obj:
print(f" ❌ 缓存对象 {j+1} 没有 size 字段!")
continue
cached_size = sampled_obj["size"]
if cached_size is None:
print(f" ❌ 缓存对象 {j+1} 的 size 为 None!")
continue
try:
size_diff = sampling_engine.calculate_size_difference(size, cached_size)
print(f" 大小差异: {size_diff} (阈值: {sampling_engine.asset_size_threshold})")
if size_diff <= sampling_engine.asset_size_threshold:
matching_obj = sampled_obj
print(f" ✅ 找到匹配的缓存对象!")
break
except Exception as e:
print(f" ❌ 计算大小差异失败: {e}")
continue
if matching_obj:
print(f" ✅ 使用缓存对象")
new_obj = copy.deepcopy(obj)
new_obj.update({
"sampled_asset_jid": matching_obj["sampled_asset_jid"],
"sampled_asset_desc": matching_obj["sampled_asset_desc"],
"sampled_asset_size": matching_obj["sampled_asset_size"],
"uuid": str(uuid.uuid4())
})
else:
print(f" 🎯 创建新的采样对象...")
new_obj = sampling_engine.create_sampled_obj(obj, batch_probs[i], True)
if new_obj is not None:
desc_size_map[desc].append(new_obj)
print(f" ✅ 新对象已添加到缓存")
else:
print(f" ❌ 新对象创建失败!")
continue
else:
print(f" 🎯 首次遇到该描述,创建新对象...")
new_obj = sampling_engine.create_sampled_obj(obj, batch_probs[i], True)
if new_obj is not None:
desc_size_map[desc] = [new_obj]
print(f" ✅ 新对象已创建缓存条目")
else:
print(f" ❌ 新对象创建失败!")
continue
if new_obj is not None:
sampled_scene["objects"].append(new_obj)
print(f" ✅ 对象已添加到最终场景")
else:
print(f" ❌ 对象为 None,跳过添加")
print(f"\n🎉 采样完成!")
print(f" 原始对象数: {len(objects)}")
print(f" 采样对象数: {len(sampled_scene['objects'])}")
print(f" 缓存条目数: {len(desc_size_map)}")
return sampled_scene
except Exception as e:
print(f"\n❌ sample_all_assets 调试过程出错:")
print(f" 错误类型: {type(e).__name__}")
print(f" 错误信息: {str(e)}")
traceback.print_exc()
return None
def main():
"""主函数"""
print("🔬 ReSpace Sample.py 调试工具")
print("专门诊断 'NoneType' object has no attribute 'get' 错误")
print("="*60)
result = debug_sample_all_assets()
print("\n" + "="*60)
if result is not None:
print("🎉 调试完成,采样过程成功执行")
print("如果之前有错误,现在应该可以看到具体的问题位置")
else:
print("❌ 调试发现错误,请查看上面的详细信息")
print("错误应该在上面的输出中有详细描述")
print("="*60)
if __name__ == "__main__":
main()