Closed junyuyang7 closed 8 months ago
notice that install mmcv-full first and make sure the version of your pytorch
pip install pytorch==1.13.1 torchvision==0.14.1 pytorch-cuda=11.7 -c pytorch -c nvidia
Apologies for the delayed response. Thanks @noahzhy for the update. @junyuyang7 did that solve the problem?
@junyuyang7 Regarding the second question, no we don't use the MS-COCO dataset for training. We convert the existing chart datasets into MS-COCO json annotation format for instance segmentation training, as that is what the MMDetection framework (that we use) natively supports.
Hope that helps :)
Hi @TheJaeLal I create a conda environment including the packages just as in your install.sh
, still got this AttributeError: 'ConfigDict' object has no attribute 'layer_cfg'
Hi @TheJaeLal I create a conda environment including the packages just as in your
install.sh
, still got thisAttributeError: 'ConfigDict' object has no attribute 'layer_cfg'
Check version of mmcv-full, it's work on my Win PC. The detail version of mm tools, check here #1
Thanks @noahzhy for the response. Closing this issue for now.
Hello .I run the code by you tutorial from Readme.md. But I got a mistake: AttributeError: 'ConfigDict' object has no attribute 'layer_cfg' Can you tell me why and how to fix it? And another question is why the paper don't mention Coco dataset, but it occur in the code. Where can I to download it?