Problems in the use of pre training model

[wudizuixiaosa]

Hello, first of all, thank you for your amazing results. I obtained the pre training model of Imagenet 1K in Mae released by Facebook research before, and then loaded my own coco format data set for training, but an error occurred. Have you encountered this type of problem before. In addition, I read your vitaev2 paper, which mentioned that you used Mae to train vitae. Is there any specific operation method, such as where to modify the overall Mae.
Traceback (most recent call last):
File "tools/train.py", line 190, in
main()
File "tools/train.py", line 186, in main
meta=meta)
File "/home/lzh/miniconda3/envs/det2/lib/python3.7/site-packages/mmdet-2.18.0-py3.7.egg/mmdet/apis/train.py", line 177, in train_detector
runner.resume(cfg.resume_from)
File "/home/lzh/miniconda3/envs/det2/lib/python3.7/site-packages/mmcv/runner/base_runner.py", line 361, in resume
self._epoch = checkpoint['meta']['epoch']
KeyError: 'meta'

[Annbless]

The pretrained models should be loaded by specifying the model.pretrained options, not the resume_from options. Please refer to Usage for detailed commands.

[wudizuixiaosa]

The pretrained models should be loaded by specifying the model.pretrained options, not the resume_from options. Please refer to Usage for detailed commands.

Ha ha, forgive me for not reading the usage carefully. Thank you for your timely reply. I wish you a happy life.

[wudizuixiaosa]

If it's convenient, can you answer by the way? I read your vitaev2 paper, which mentioned that you train vitae with MAE. Whether there are any specific operation methods, such as where to modify the overall MAE. I am very interested in this because Mae cannot use PVT or swin due to its specific input method.

[Annbless]

Hi,

Please refer to Sec 3.3 and Sec 4.4.1 in the ViTAEv2 paper. We use MAE to train the isotropic design of ViTAE, not the hierarchical ViTAEv2 backbone. We will explore training hierarchical ViTAEv2 backbone using MIM series methods in the future.

[xmdgaoxin]

Hello, how can I solve the following error when I use my own coco dataset during training