数据预处理过程内存占用过高以至报错

[FirokOtaku]

执行 python tools/preprocess.py --config-file configs/preprocess/dota1_5_preprocess_config.py 时,
内存占用量在命令开始执行之后若干秒便开始激增,
直到吃满所有系统内存.

系统环境:

• Windows 专业版 21H1 19043.1706
• Python 3.10.7
• 内存 32G
• 显卡 3070

开始部分日志:

(数据集目录名是 dota-1.0, 里面的数据使用的是 1.5 版本)

D:\project-jittor\jdet>python tools\preprocess.py --config-file configs/preprocess/dota1_5_preprocess_config.py
[i 0930 15:16:14.694000 84 compiler.py:955] Jittor(1.3.5.16) src: c:\users\pc\appdata\local\programs\python\python310\lib\site-packages\jittor-1.3.5.16-py3.10.egg\jittor
[i 0930 15:16:14.740000 84 compiler.py:956] cl at C:\Users\pc\.cache\jittor\msvc\VC\_\_\_\_\_\bin\cl.exe(19.29.30133)
[i 0930 15:16:14.741000 84 compiler.py:957] cache_path: C:\Users\pc\.cache\jittor\jt1.3.5\cl\py3.10.7\Windows-10-10.xb1\IntelRCoreTMi7x95\default
[i 0930 15:16:14.745000 84 install_cuda.py:88] cuda_driver_version: [11, 7, 0]
[i 0930 15:16:14.796000 84 __init__.py:411] Found C:\Users\pc\.cache\jittor\jtcuda\cuda11.2_cudnn8_win\bin\nvcc.exe(11.2.67) at C:\Users\pc\.cache\jittor\jtcuda\cuda11.2_cudnn8_win\bin\nvcc.exe.
[i 0930 15:16:14.911000 84 compiler.py:1010] cuda key:cu11.2.67
[i 0930 15:16:14.913000 84 __init__.py:227] Total mem: 31.91GB, using 10 procs for compiling.
[i 0930 15:16:15.796000 84 jit_compiler.cc:28] Load cc_path: C:\Users\pc\.cache\jittor\msvc\VC\_\_\_\_\_\bin\cl.exe
[i 0930 15:16:15.797000 84 init.cc:62] Found cuda archs: [86,]
[i 0930 15:16:15.892000 84 compile_extern.py:517] mpicc not found, distribution disabled.
[w 0930 15:16:15.941000 84 compile_extern.py:200] CUDA related path found in LD_LIBRARY_PATH or PATH(['', 'C', '\\Users\\pc\\.cache\\jittor\\jtcuda\\cuda11.2_cudnn8_win\\lib64', '', 'C', '\\Users\\pc\\.cache\\jittor\\mkl\\dnnl_win_2.2.0_cpu_vcomp\\bin', '', 'C', '\\Users\\pc\\.cache\\jittor\\mkl\\dnnl_win_2.2.0_cpu_vcomp\\lib', '', 'C', '\\Users\\pc\\.cache\\jittor\\jt1.3.5\\cl\\py3.10.7\\Windows-10-10.xb1\\IntelRCoreTMi7x95\\default', '', 'C', '\\Users\\pc\\.cache\\jittor\\jt1.3.5\\cl\\py3.10.7\\Windows-10-10.xb1\\IntelRCoreTMi7x95\\default\\cu11.2.67', '', 'C', '\\Users\\pc\\.cache\\jittor\\jtcuda\\cuda11.2_cudnn8_win\\bin', '', 'C', '\\Users\\pc\\.cache\\jittor\\jtcuda\\cuda11.2_cudnn8_win\\lib\\x64', '', 'C', '\\Users\\pc\\.cache\\jittor\\msvc\\win10_kits\\lib\\ucrt\\x64', '', 'C', '\\Users\\pc\\.cache\\jittor\\msvc\\win10_kits\\lib\\um\\x64', '', 'C', '\\Users\\pc\\.cache\\jittor\\msvc\\VC\\lib', '', 'c', '\\users\\pc\\appdata\\local\\programs\\python\\python310\\libs', 'C', '\\Users\\pc\\.cache\\jittor\\msvc\\VC\\_\\_\\_\\_\\_\\bin', 'C', '\\Users\\pc\\AppData\\Local\\Programs\\Python\\Python310\\Lib\\site-packages\\opencv_python-4.6.0.66-py3.10-win-amd64.egg\\cv2\\../../x64/vc14/bin', 'C', '\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v11.7\\bin', 'C', '\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v11.7\\libnvvp', 'D', '\\Release', 'D', '\\ffmpeg\\bin', 'C', '\\ProgramData\\Oracle\\Java\\javapath', 'C', '\\Program Files\\Java\\jdk1.8.0_131\\bin', 'C', '\\Program Files\\Java\\jdk1.8.0_131\\jre\\bin', 'C', '\\Windows\\system32', 'C', '\\Windows', 'C', '\\Windows\\System32\\Wbem', 'C', '\\Windows\\System32\\WindowsPowerShell\\v1.0\\', 'C', '\\Windows\\System32\\OpenSSH\\', 'C', '\\Program Files (x86)\\NVIDIA Corporation\\PhysX\\Common', 'C', '\\Program Files\\NVIDIA Corporation\\NVIDIA NvDLISR', 'C', '\\Program Files\\NVIDIA Corporation\\Nsight Compute 2022.2.1\\', 'C', '\\Users\\pc\\AppData\\Local\\Programs\\Python\\Python310\\Scripts\\', 'C', '\\Users\\pc\\AppData\\Local\\Programs\\Python\\Python310\\', 'C', '\\Program Files\\MySQL\\MySQL Shell 8.0\\bin\\', 'C', '\\Users\\pc\\AppData\\Local\\Microsoft\\WindowsApps', '', 'C', '\\Users\\pc\\AppData\\Local\\Programs\\Microsoft VS Code\\bin']), This path may cause jittor found the wrong libs, please unset LD_LIBRARY_PATH and remove cuda lib path in Path.
Or you can let jittor install cuda for you: `python3.x -m jittor_utils.install_cuda`
Loading config from:  configs/preprocess/dota1_5_preprocess_config.py
{'type': 'DOTA1_5', 'source_dataset_path': 'D:\\project-jittor\\dataset\\dota-1.0', 'target_dataset_path': 'D:\\project-jittor\\dataset\\dota-1.0-processed', 'tasks': [{'label': 'trainval', 'config': {'subimage_size': 600, 'overlap_size': 150, 'multi_scale': [1.0], 'horizontal_flip': False, 'vertical_flip': False, 'rotation_angles': [0.0]}}, {'label': 'test', 'config': {'subimage_size': 600, 'overlap_size': 150, 'multi_scale': [1.0], 'horizontal_flip': False, 'vertical_flip': False, 'rotation_angles': [0.0]}}], 'name': 'dota1_5_preprocess_config', 'work_dir': 'work_dirs/dota1_5_preprocess_config'}
==============
processing trainval
fatal   fatal   : : Memory allocation failureMemory allocation failure

fatal   : Memory allocation failure
系统无法执行指定的程序。
内存资源不足，无法处理此命令。
内存资源不足，无法处理此命令。
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "c:\users\pc\appdata\local\programs\python\python310\lib\multiprocessing\spawn.py", line 116, in spawn_main
    exitcode = _main(fd, parent_sentinel)
  File "c:\users\pc\appdata\local\programs\python\python310\lib\multiprocessing\spawn.py", line 125, in _main
    prepare(preparation_data)
  File "c:\users\pc\appdata\local\programs\python\python310\lib\multiprocessing\spawn.py", line 236, in prepare
    _fixup_main_from_path(data['init_main_from_path'])
  File "c:\users\pc\appdata\local\programs\python\python310\lib\multiprocessing\spawn.py", line 287, in _fixup_main_from_path
    main_content = runpy.run_path(main_path,
  File "c:\users\pc\appdata\local\programs\python\python310\lib\runpy.py", line 289, in run_path
    return _run_module_code(code, init_globals, run_name,
  File "c:\users\pc\appdata\local\programs\python\python310\lib\runpy.py", line 96, in _run_module_code
    _run_code(code, mod_globals, init_globals,
  File "c:\users\pc\appdata\local\programs\python\python310\lib\runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "D:\project-jittor\jdet\tools\preprocess.py", line 5, in <module>
    from jdet.config import init_cfg, get_cfg
  File "d:\project-jittor\jdet\python\jdet\__init__.py", line 1, in <module>
    from . import models
  File "d:\project-jittor\jdet\python\jdet\models\__init__.py", line 1, in <module>
    from .networks import *
  File "d:\project-jittor\jdet\python\jdet\models\networks\__init__.py", line 1, in <module>
    from .rcnn import RCNN
  File "d:\project-jittor\jdet\python\jdet\models\networks\rcnn.py", line 2, in <module>
    import jittor as jt
  File "C:\Users\pc\AppData\Local\Programs\Python\Python310\lib\site-packages\jittor-1.3.5.16-py3.10.egg\jittor\__init__.py", line 32, in <module>
    from typing import List, Tuple
  File "C:\Users\pc\AppData\Local\Programs\Python\Python310\lib\typing.py", line 2178, in <module>
    class SupportsInt(Protocol):
  File "c:\users\pc\appdata\local\programs\python\python310\lib\abc.py", line 106, in __new__
    cls = super().__new__(mcls, name, bases, namespace, **kwargs)
  File "C:\Users\pc\AppData\Local\Programs\Python\Python310\lib\typing.py", line 1554, in __init_subclass__
    cls._is_protocol = any(b is Protocol for b in cls.__bases__)
MemoryError: Out of memory interning an attribute name
1 / 10
/ 0
1 / 0
C:\Users\pc\AppData\Local\Programs\Python\Python310\lib\site-packages\shapely-2.0a1-py3.10-win-amd64.egg\shapely\set_operations.py:132: RuntimeWarning: invalid value encountered in intersection
  return lib.intersection(a, b, **kwargs)
C:\Users\pc\AppData\Local\Programs\Python\Python310\lib\site-packages\shapely-2.0a1-py3.10-win-amd64.egg\shapely\set_operations.py:132: RuntimeWarning: invalid value encountered in intersection
  return lib.intersection(a, b, **kwargs)
C:\Users\pc\AppData\Local\Programs\Python\Python310\lib\site-packages\shapely-2.0a1-py3.10-win-amd64.egg\shapely\set_operations.py:132: RuntimeWarning: invalid value encountered in intersection
  return lib.intersection(a, b, **kwargs)
2 / 0
2 / 0
3 / 0
4 / 0

在这后面有很多类似的 "数字 / 0" 日志, 以及下面这样的日志:

C:\Users\pc\AppData\Local\Programs\Python\Python310\lib\site-packages\shapely-2.0a1-py3.10-win-amd64.egg\shapely\set_operations.py:132: RuntimeWarning: invalid value encountered in intersection
  return lib.intersection(a, b, **kwargs)

命令执行过程中会弹出系统错误弹窗: nvcc.exe 应用程序无法正常启动(0xc0000142).请单击"确定"关闭应用程序.

再往后有大量类似的报错日志:

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "c:\users\pc\appdata\local\programs\python\python310\lib\multiprocessing\spawn.py", line 116, in spawn_main
    exitcode = _main(fd, parent_sentinel)
  File "c:\users\pc\appdata\local\programs\python\python310\lib\multiprocessing\spawn.py", line 125, in _main
    prepare(preparation_data)
  File "c:\users\pc\appdata\local\programs\python\python310\lib\multiprocessing\spawn.py", line 236, in prepare
Traceback (most recent call last):
      File "<string>", line 1, in <module>
_fixup_main_from_path(data['init_main_from_path'])
  File "c:\users\pc\appdata\local\programs\python\python310\lib\multiprocessing\spawn.py", line 287, in _fixup_main_from_path
    main_content = runpy.run_path(main_path,
  File "c:\users\pc\appdata\local\programs\python\python310\lib\runpy.py", line 289, in run_path
  File "c:\users\pc\appdata\local\programs\python\python310\lib\multiprocessing\spawn.py", line 116, in spawn_main
    return _run_module_code(code, init_globals, run_name,
exitcode = _main(fd, parent_sentinel)  File "c:\users\pc\appdata\local\programs\python\python310\lib\runpy.py", line 96, in _run_module_code

      File "c:\users\pc\appdata\local\programs\python\python310\lib\multiprocessing\spawn.py", line 125, in _main
_run_code(code, mod_globals, init_globals,
prepare(preparation_data)  File "c:\users\pc\appdata\local\programs\python\python310\lib\runpy.py", line 86, in _run_code

      File "c:\users\pc\appdata\local\programs\python\python310\lib\multiprocessing\spawn.py", line 236, in prepare
exec(code, run_globals)
_fixup_main_from_path(data['init_main_from_path'])  File "D:\project-jittor\jdet\tools\preprocess.py", line 5, in <module>

      File "c:\users\pc\appdata\local\programs\python\python310\lib\multiprocessing\spawn.py", line 287, in _fixup_main_from_path
from jdet.config import init_cfg, get_cfg
      File "d:\project-jittor\jdet\python\jdet\__init__.py", line 1, in <module>
main_content = runpy.run_path(main_path,
from . import models  File "c:\users\pc\appdata\local\programs\python\python310\lib\runpy.py", line 289, in run_path

  File "d:\project-jittor\jdet\python\jdet\models\__init__.py", line 1, in <module>
    return _run_module_code(code, init_globals, run_name,
  File "c:\users\pc\appdata\local\programs\python\python310\lib\runpy.py", line 96, in _run_module_code
    _run_code(code, mod_globals, init_globals,
  File "c:\users\pc\appdata\local\programs\python\python310\lib\runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "D:\project-jittor\jdet\tools\preprocess.py", line 5, in <module>
    from jdet.config import init_cfg, get_cfg
  File "d:\project-jittor\jdet\python\jdet\__init__.py", line 1, in <module>
    from . import models
  File "d:\project-jittor\jdet\python\jdet\models\__init__.py", line 1, in <module>
        from .networks import *from .networks import *

  File "d:\project-jittor\jdet\python\jdet\models\networks\__init__.py", line 1, in <module>
  File "d:\project-jittor\jdet\python\jdet\models\networks\__init__.py", line 1, in <module>

MemoryError: Out of memory interning an attribute name

ImportError: DLL load failed while importing cv2: 页面文件太小，无法完成操作。

请问如此高的内存占用是否是正常情况?
有无限制内存使用量的配置项?
还是说目前只能使用更高内存的设备运行预处理指令?

[FirokOtaku]

已在讨论群找到解决方案:
修改 jdet/python/jdet/data/devkits/ImgSplit_multi_process.py 中位于 329 行的 num_process 值,
将默认的 32 调低即可.

在我这边的环境下调到 8 就没什么问题了,
可以正常执行完成预处理脚本.