Compatibility Issue with MMCV, PyTorch, and CUDA in Google Colab Object Detection Model

[faizan628]

Thanks for your error report and we appreciate it a lot.

Checklist

1. I have searched related issues but cannot get the expected help.
2. I have read the FAQ documentation but cannot get the expected help.
3. The bug has not been fixed in the latest version and previous versions

Describe the bug
I am encountering a compatibility issue between MMCV, PyTorch, and the CUDA GPU environment while training a model for object detection using Google Colab. The same code was working perfectly fine last year, but now it fails. The dataset is confirmed to be correct, as I used it successfully for training in the past. However, the model training fails now.

Reproduction

from google.colab import drive
drive.mount('/content/drive')

!cp -r '/content/drive/MyDrive/Table_detection_paper/' '/content/'

Model configurations

MODELS_CONFIG = {
'faster_rcnn_r50_fpn_1x': {
'config_file': 'configs/pascal_voc/faster-rcnn_r50_fpn_1x_voc0712.py'
},
'cascade_rcnn_r50_fpn_1x': {
'config_file': 'configs/cascade_rcnn_r50_fpn_1x.py',
},
'retinanet_r50_fpn_1x': {
'config_file': 'configs/retinanet_r50_fpn_1x.py',
}
}

mmdetection root directory

root_dir = '/content/finance_research'

Pick the model you want to use

Select a model in MODELS_CONFIG.

selected_model = 'faster_rcnn_r50_fpn_1x' # 'faster_rcnn_r50_fpn_1x' # cascade_rcnn_r50_fpn_1x

Total training epochs.

total_epochs = 12

Name of the config file.

config_file = MODELS_CONFIG[selected_model]['config_file']

import os
from os.path import exists, join, basename, splitext

%cd /content
project_name = os.path.abspath(splitext(basename(root_dir))[0])
%cd {project_name}
mmdetection_dir = os.path.join(project_name, "mmdetection")

!pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118

!pip install -U openmim

!mim install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu118/
!pip3 install openmim
!mim install mmengine

!git clone --branch v3.2.0 https://github.com/open-mmlab/mmdetection.git
%cd mmdetection
!pip install -v -e .

%cd {mmdetection_dir}
!python tools/train.py {config_fname}

A placeholder for the command.

2. Did you make any modifications on the code or config? Did you understand what you have modified?
3. What dataset did you use?

Environment

Google Colab
PyTorch version: 2.0.0
Torchvision version: 0.15.1
CUDA version: cu118
MMCV version: 1.7.2 (incompatible)
MMDetection version: 3.2.0
Error traceback
/usr/local/lib/python3.10/dist-packages/mmcv/init.py:20: UserWarning: On January 1, 2023, MMCV will release v2.0.0, in which it will remove components related to the training process and add a data transformation module. In addition, it will rename the package names mmcv to mmcv-lite and mmcv-full to mmcv.
Traceback (most recent call last):
File "/content/finance_research/mmdetection/tools/train.py", line 10, in
from mmdet.utils import setup_cache_size_limit_of_dynamo
File "/content/finance_research/mmdetection/mmdet/init.py", line 17, in
and mmcv_version < digit_version(mmcv_maximum_version)),
AssertionError: MMCV==1.7.2 is used but incompatible. Please install mmcv>=2.0.0rc4, <2.2.0.

A placeholder for trackback.

Bug fix
If you have already identified the reason, you can provide the information here. If you are willing to create a PR to fix it, please also leave a comment here and that would be much appreciated!

[faizan628]

Hi 王宇东，

希望你一切顺利。你能否帮我解决我遇到的问题？我非常感激你的支持。

如果方便的话，我们能否安排一次通话，以便更好地理解和讨论这个问题？

此致，
法伊赞·西迪奇

[shallweiwei]

you can pip mmcv>=2.0.0